@nodaro/shared 3.60.1 → 3.61.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/{chunk-DS2TOM32.js → chunk-43TGSZRJ.js} +27 -3
- package/dist/chunk-43TGSZRJ.js.map +1 -0
- package/dist/{chunk-RRPPXZ36.cjs → chunk-4KYLNTR7.cjs} +4 -4
- package/dist/{chunk-RRPPXZ36.cjs.map → chunk-4KYLNTR7.cjs.map} +1 -1
- package/dist/{chunk-HDHLZK7C.cjs → chunk-5I4T4QR5.cjs} +5 -5
- package/dist/{chunk-HDHLZK7C.cjs.map → chunk-5I4T4QR5.cjs.map} +1 -1
- package/dist/{chunk-DQIJPDHC.cjs → chunk-5XRRENHD.cjs} +11 -11
- package/dist/{chunk-DQIJPDHC.cjs.map → chunk-5XRRENHD.cjs.map} +1 -1
- package/dist/{chunk-LN3EVFRS.js → chunk-673E4KYC.js} +5 -5
- package/dist/{chunk-LN3EVFRS.js.map → chunk-673E4KYC.js.map} +1 -1
- package/dist/{chunk-CIOXUPCJ.cjs → chunk-7ARKVJTN.cjs} +12 -12
- package/dist/{chunk-CIOXUPCJ.cjs.map → chunk-7ARKVJTN.cjs.map} +1 -1
- package/dist/{chunk-MHY47BLL.js → chunk-7WGKP3EU.js} +4 -4
- package/dist/{chunk-MHY47BLL.js.map → chunk-7WGKP3EU.js.map} +1 -1
- package/dist/{chunk-KOLC7AF2.js → chunk-A3FHTSLB.js} +99 -14
- package/dist/chunk-A3FHTSLB.js.map +1 -0
- package/dist/{chunk-6Q6SI7AG.js → chunk-B6BG3LV2.js} +3 -3
- package/dist/{chunk-6Q6SI7AG.js.map → chunk-B6BG3LV2.js.map} +1 -1
- package/dist/{chunk-SCYBGTX6.js → chunk-C77F4ZTL.js} +3 -3
- package/dist/{chunk-SCYBGTX6.js.map → chunk-C77F4ZTL.js.map} +1 -1
- package/dist/{chunk-IV6IDECY.js → chunk-EWS2VFO4.js} +4 -4
- package/dist/{chunk-IV6IDECY.js.map → chunk-EWS2VFO4.js.map} +1 -1
- package/dist/{chunk-A6HRY3WF.cjs → chunk-F5YEMSNK.cjs} +102 -13
- package/dist/chunk-F5YEMSNK.cjs.map +1 -0
- package/dist/{chunk-5HK5Z5OB.cjs → chunk-FMXFHJWX.cjs} +4 -4
- package/dist/{chunk-5HK5Z5OB.cjs.map → chunk-FMXFHJWX.cjs.map} +1 -1
- package/dist/{chunk-GWBQ3GPO.cjs → chunk-FQXRYPCX.cjs} +27 -22
- package/dist/chunk-FQXRYPCX.cjs.map +1 -0
- package/dist/{chunk-ICDLYU4I.js → chunk-HGI2VKIV.js} +8 -8
- package/dist/{chunk-ICDLYU4I.js.map → chunk-HGI2VKIV.js.map} +1 -1
- package/dist/{chunk-EN2YI5TV.js → chunk-IRQ3SPX7.js} +3 -3
- package/dist/{chunk-EN2YI5TV.js.map → chunk-IRQ3SPX7.js.map} +1 -1
- package/dist/{chunk-5Y42RIGY.cjs → chunk-IVXSNW2B.cjs} +8 -8
- package/dist/{chunk-5Y42RIGY.cjs.map → chunk-IVXSNW2B.cjs.map} +1 -1
- package/dist/{chunk-W4M7TNXG.cjs → chunk-JHTT2V4A.cjs} +39 -39
- package/dist/{chunk-W4M7TNXG.cjs.map → chunk-JHTT2V4A.cjs.map} +1 -1
- package/dist/{chunk-VYQ25YQF.js → chunk-KHZAQGRP.js} +3 -3
- package/dist/{chunk-VYQ25YQF.js.map → chunk-KHZAQGRP.js.map} +1 -1
- package/dist/{chunk-G6GMY5BC.js → chunk-LLOIAGFB.js} +9 -4
- package/dist/chunk-LLOIAGFB.js.map +1 -0
- package/dist/{chunk-AZGONXWV.js → chunk-MOMFX32D.js} +4 -4
- package/dist/{chunk-AZGONXWV.js.map → chunk-MOMFX32D.js.map} +1 -1
- package/dist/{chunk-BXNPO3PL.cjs → chunk-MQZRQ5LM.cjs} +8 -8
- package/dist/{chunk-BXNPO3PL.cjs.map → chunk-MQZRQ5LM.cjs.map} +1 -1
- package/dist/chunk-NCOGGRRP.cjs +27 -0
- package/dist/{chunk-G7ZYZQKO.cjs.map → chunk-NCOGGRRP.cjs.map} +1 -1
- package/dist/{chunk-AXT7YNOR.cjs → chunk-NJVOJXBS.cjs} +43 -43
- package/dist/{chunk-AXT7YNOR.cjs.map → chunk-NJVOJXBS.cjs.map} +1 -1
- package/dist/{chunk-SKFJIFEV.cjs → chunk-RQ4F3AJH.cjs} +21 -21
- package/dist/{chunk-SKFJIFEV.cjs.map → chunk-RQ4F3AJH.cjs.map} +1 -1
- package/dist/{chunk-HYBACFYW.js → chunk-RUHKBEFF.js} +4 -4
- package/dist/{chunk-HYBACFYW.js.map → chunk-RUHKBEFF.js.map} +1 -1
- package/dist/{chunk-URLDVK67.js → chunk-SCMTN57Y.js} +3 -3
- package/dist/{chunk-URLDVK67.js.map → chunk-SCMTN57Y.js.map} +1 -1
- package/dist/{chunk-JKZVRKN3.js → chunk-UARGJ5IZ.js} +6 -6
- package/dist/chunk-UARGJ5IZ.js.map +1 -0
- package/dist/{chunk-6AJOTEVC.js → chunk-VOWJCUMR.js} +3 -3
- package/dist/{chunk-6AJOTEVC.js.map → chunk-VOWJCUMR.js.map} +1 -1
- package/dist/{chunk-5G23AJBU.cjs → chunk-XUJC6GHE.cjs} +6 -6
- package/dist/{chunk-5G23AJBU.cjs.map → chunk-XUJC6GHE.cjs.map} +1 -1
- package/dist/{chunk-GNPJ2JLE.cjs → chunk-XZKFZMRT.cjs} +6 -6
- package/dist/{chunk-GNPJ2JLE.cjs.map → chunk-XZKFZMRT.cjs.map} +1 -1
- package/dist/{chunk-KJZVQIN4.js → chunk-ZDBPQBAA.js} +3 -3
- package/dist/{chunk-KJZVQIN4.js.map → chunk-ZDBPQBAA.js.map} +1 -1
- package/dist/{chunk-HOKWJXHA.cjs → chunk-ZFJN23V4.cjs} +29 -2
- package/dist/chunk-ZFJN23V4.cjs.map +1 -0
- package/dist/{chunk-HNEF2N5C.cjs → chunk-ZLDFPVVI.cjs} +5 -5
- package/dist/{chunk-HNEF2N5C.cjs.map → chunk-ZLDFPVVI.cjs.map} +1 -1
- package/dist/{chunk-BDXXOEBZ.cjs → chunk-ZNM2LW3D.cjs} +47 -47
- package/dist/chunk-ZNM2LW3D.cjs.map +1 -0
- package/dist/{chunk-YHVEZVAO.js → chunk-ZZGQRRN7.js} +4 -4
- package/dist/{chunk-YHVEZVAO.js.map → chunk-ZZGQRRN7.js.map} +1 -1
- package/dist/entries/apps.cjs +42 -42
- package/dist/entries/apps.js +7 -7
- package/dist/entries/credits.cjs +151 -151
- package/dist/entries/credits.d.cts +8 -4
- package/dist/entries/credits.d.ts +8 -4
- package/dist/entries/credits.js +7 -7
- package/dist/entries/llm-models.d.cts +1 -1
- package/dist/entries/llm-models.d.ts +1 -1
- package/dist/entries/model-catalog.cjs +94 -78
- package/dist/entries/model-catalog.d.cts +2 -2
- package/dist/entries/model-catalog.d.ts +2 -2
- package/dist/entries/model-catalog.js +4 -4
- package/dist/entries/models.cjs +293 -293
- package/dist/entries/models.d.cts +3 -3
- package/dist/entries/models.d.ts +3 -3
- package/dist/entries/models.js +4 -4
- package/dist/entries/nodes.cjs +206 -206
- package/dist/entries/nodes.d.cts +1 -1
- package/dist/entries/nodes.d.ts +1 -1
- package/dist/entries/nodes.js +8 -8
- package/dist/entries/pipeline.cjs +152 -152
- package/dist/entries/pipeline.d.cts +1 -1
- package/dist/entries/pipeline.d.ts +1 -1
- package/dist/entries/pipeline.js +3 -3
- package/dist/entries/ugc.cjs +71 -71
- package/dist/entries/ugc.js +4 -4
- package/dist/entries/video.cjs +117 -117
- package/dist/entries/video.js +4 -4
- package/dist/entries/workflow.cjs +110 -98
- package/dist/entries/workflow.d.cts +28 -2
- package/dist/entries/workflow.d.ts +28 -2
- package/dist/entries/workflow.js +1 -1
- package/dist/index.cjs +1212 -1184
- package/dist/index.d.cts +3 -3
- package/dist/index.d.ts +3 -3
- package/dist/index.js +18 -18
- package/dist/{model-catalog-D6Es7VG-.d.ts → model-catalog-BgSvskgY.d.cts} +52 -1
- package/dist/{model-catalog-D6Es7VG-.d.cts → model-catalog-BgSvskgY.d.ts} +52 -1
- package/dist/{model-constants-22k3fjp1.d.cts → model-constants-DHd9jwQ7.d.cts} +6 -1
- package/dist/{model-constants-22k3fjp1.d.ts → model-constants-DHd9jwQ7.d.ts} +6 -1
- package/package.json +1 -1
- package/src/__tests__/gpt-image-2-6.test.ts +7 -0
- package/src/__tests__/image-mask-mode.test.ts +34 -1
- package/src/__tests__/mask-edit-follows-source.test.ts +150 -0
- package/src/credit-identifiers.ts +8 -4
- package/src/entries/model-catalog.ts +4 -0
- package/src/entries/workflow.ts +1 -0
- package/src/model-catalog.ts +157 -11
- package/src/model-constants.ts +12 -2
- package/src/node-execution-state.ts +3 -1
- package/src/wired-image-size.ts +55 -0
- package/dist/chunk-A6HRY3WF.cjs.map +0 -1
- package/dist/chunk-BDXXOEBZ.cjs.map +0 -1
- package/dist/chunk-DS2TOM32.js.map +0 -1
- package/dist/chunk-G6GMY5BC.js.map +0 -1
- package/dist/chunk-G7ZYZQKO.cjs +0 -27
- package/dist/chunk-GWBQ3GPO.cjs.map +0 -1
- package/dist/chunk-HOKWJXHA.cjs.map +0 -1
- package/dist/chunk-JKZVRKN3.js.map +0 -1
- package/dist/chunk-KOLC7AF2.js.map +0 -1
|
@@ -310,6 +310,21 @@ interface ModelCatalogEntry {
|
|
|
310
310
|
* as before. "auto" stays in `aspectRatios`, so every menu still offers it.
|
|
311
311
|
*/
|
|
312
312
|
autoAspectFromSourceImage?: true;
|
|
313
|
+
/**
|
|
314
|
+
* The pixel size (width, height) this model renders at each aspect ratio and
|
|
315
|
+
* size, from the provider's own table, for a PNG, JPEG or WebP. Read where a
|
|
316
|
+
* run's size is chosen from its photo's size (`maskEditMatchesSource`).
|
|
317
|
+
*/
|
|
318
|
+
outputSizes?: Readonly<Record<string, Readonly<Record<string, readonly [number, number]>>>>;
|
|
319
|
+
/**
|
|
320
|
+
* A masked edit on this model follows its photo: the result is pasted back
|
|
321
|
+
* onto the photo at the photo's own size, so the run keeps the photo's shape
|
|
322
|
+
* ("auto", whatever ratio it names) and renders at the size of `outputSizes`
|
|
323
|
+
* whose long side is nearest the photo's, priced at that size
|
|
324
|
+
* (`normalizeModelInput` with `context.masked`; `resolutionNearestSource`).
|
|
325
|
+
* With the photo's size unknown, the size asked is kept.
|
|
326
|
+
*/
|
|
327
|
+
maskEditMatchesSource?: true;
|
|
313
328
|
resolutions?: readonly string[];
|
|
314
329
|
/**
|
|
315
330
|
* The aspect ratios at which this model renders fewer sizes than
|
|
@@ -591,9 +606,45 @@ interface ModelInputContext {
|
|
|
591
606
|
* resolve "auto" on a model without a native auto, or whose own auto does
|
|
592
607
|
* not keep the photo's shape (`autoAspectFromSourceImage`); absent or
|
|
593
608
|
* unusable means the size is unknown and "auto" snaps the way it always has.
|
|
609
|
+
* Also read by a masked edit on a model whose result follows its photo
|
|
610
|
+
* (`maskEditMatchesSource`): it decides the ratio and the size.
|
|
594
611
|
*/
|
|
595
612
|
sourceImage?: SourceImageSize;
|
|
613
|
+
/** The request carries an inpainting mask (`maskUrl`). */
|
|
614
|
+
masked?: boolean;
|
|
596
615
|
}
|
|
616
|
+
/**
|
|
617
|
+
* True when a masked request on `modelId` follows its photo
|
|
618
|
+
* (`maskEditMatchesSource`): its ratio and size come from the photo's size.
|
|
619
|
+
*/
|
|
620
|
+
declare function maskEditMatchesSource(modelId: string | undefined, masked: unknown): boolean;
|
|
621
|
+
/**
|
|
622
|
+
* True when the photo's size can change what a request on `modelId` sends:
|
|
623
|
+
* "auto" that keeps the photo's shape (`autoAspectNeedsSourceImage`), or a
|
|
624
|
+
* masked edit that follows its photo (`maskEditMatchesSource`). A caller that
|
|
625
|
+
* has to fetch the size asks this first.
|
|
626
|
+
*/
|
|
627
|
+
declare function sourceImageSizeNeeded(modelId: string | undefined, aspectRatio: unknown, masked: unknown): boolean;
|
|
628
|
+
/**
|
|
629
|
+
* The size a MASKED request on `modelId` runs, and is priced, at once its
|
|
630
|
+
* photo's size is known — what `normalizeModelInput` sends for it. For an
|
|
631
|
+
* estimate that knows the photo's size (the editor's badge, a workflow
|
|
632
|
+
* estimate). `undefined` when the model does not follow its photo
|
|
633
|
+
* (`maskEditMatchesSource`) or the size is unusable: the caller keeps its own.
|
|
634
|
+
*/
|
|
635
|
+
declare function maskedEditResolution(modelId: string, levers: {
|
|
636
|
+
aspectRatio?: unknown;
|
|
637
|
+
resolution?: unknown;
|
|
638
|
+
background?: unknown;
|
|
639
|
+
outputFormat?: unknown;
|
|
640
|
+
}, sourceImage: SourceImageSize | undefined): string | undefined;
|
|
641
|
+
/**
|
|
642
|
+
* The size of `modelId` whose rendered long side, at `aspectRatio`, is nearest
|
|
643
|
+
* the photo's long side (`outputSizes`; nearest by ratio, so 1500 px is nearer
|
|
644
|
+
* 2048 than 1024; a tie goes to the larger size). `undefined` when the model
|
|
645
|
+
* has no size table for that ratio or the photo's size is unusable.
|
|
646
|
+
*/
|
|
647
|
+
declare function resolutionNearestSource(modelId: string, aspectRatio: unknown, source: SourceImageSize | undefined): string | undefined;
|
|
597
648
|
/** True for the "auto" aspect token in any case or padding. */
|
|
598
649
|
declare function isAutoAspectToken(value: unknown): boolean;
|
|
599
650
|
/**
|
|
@@ -833,4 +884,4 @@ declare function modelIdsByKindMode(kind: ModelKind | null, modes: readonly Mode
|
|
|
833
884
|
includeHidden?: boolean;
|
|
834
885
|
}): [string, ...string[]];
|
|
835
886
|
|
|
836
|
-
export {
|
|
887
|
+
export { modelMakerOf as $, creditRangesAll as A, defaultResolutionFor as B, durationsByMode as C, fitAspectRatioToModel as D, getAspectRatioOptions as E, getCreditRange as F, getDurationsForModel as G, getModel as H, IMAGE_GIF_FRAME_SIZE as I, getQualityOptions as J, getResolutionOptions as K, type LabeledOption as L, MASKLESS_OUTPUT_FORMATS as M, type NormalizedModelInput as N, groupByFamily as O, type PriceVariant as P, groupByKindAndFamily as Q, hasFeature as R, SPEECH_UNIT_PRICE_NOTE as S, TRANSPARENT_OUTPUT_FORMATS as T, isAutoAspectToken as U, type ValidationField as V, isModelMakerId as W, listModels as X, maskEditMatchesSource as Y, maskedEditResolution as Z, modelIdsByKindMode as _, IMAGE_OUTPUT_FORMAT_LABELS as a, modelTakesSeed as a0, modelsWithFeature as a1, nearestCatalogAspectRatio as a2, normalizeModelInput as a3, normalizeVideoRequestParams as a4, outputFormatOptionsByKind as a5, qualityOptionsByKind as a6, resolutionAtAspectRatio as a7, resolutionNearestSource as a8, resolutionOptionsByKind as a9, resolutionsAtAspectRatio as aa, sourceImageSizeNeeded as ab, svgLevelOptionsByKind as ac, validateModelInput as ad, IMAGE_SVG_LEVEL_LABELS as b, MODEL_CATALOG as c, MODEL_INPUT_ADJUSTMENT_FIELDS as d, MODEL_KINDS as e, MODEL_MAKERS as f, MODEL_RECOMMENDATIONS as g, type ModelCatalogEntry as h, type ModelInputAdjustment as i, type ModelInputAdjustmentField as j, type ModelInputContext as k, type ModelKind as l, type ModelMaker as m, type ModelMakerId as n, type ModelMenuOption as o, type ModelMode as p, type ModelRecommendation as q, type ModelValidationIssue as r, type NormalizedVideoRequest as s, type SourceImageSize as t, type TtsCapabilities as u, type TtsSettingLever as v, aspectRatioOptionsByKind as w, autoAspectNeedsSourceImage as x, backgroundOptionsByKind as y, buildModelMenu as z };
|
|
@@ -823,7 +823,12 @@ type VoiceDesignModel = typeof VOICE_DESIGN_MODELS[number];
|
|
|
823
823
|
* (`backend/src/providers/elevenlabs/voice-design.ts`) can't drift.
|
|
824
824
|
*/
|
|
825
825
|
declare const DEFAULT_VOICE_DESIGN_MODEL: VoiceDesignModel;
|
|
826
|
-
/**
|
|
826
|
+
/**
|
|
827
|
+
* I2I providers that take an inpainting mask (`maskUrl`, white = the area to
|
|
828
|
+
* change, black = keep) as the model's own mask. The provider converts the
|
|
829
|
+
* mask into the form its model reads. Every catalog entry listing the
|
|
830
|
+
* `"mask"` feature (`/v1/models`, MCP `list_models`) is one of these.
|
|
831
|
+
*/
|
|
827
832
|
declare const I2I_MASK_SUPPORT: Set<string>;
|
|
828
833
|
/**
|
|
829
834
|
* Mask edit tier per image-gen provider (single source of truth for inpaint).
|
|
@@ -823,7 +823,12 @@ type VoiceDesignModel = typeof VOICE_DESIGN_MODELS[number];
|
|
|
823
823
|
* (`backend/src/providers/elevenlabs/voice-design.ts`) can't drift.
|
|
824
824
|
*/
|
|
825
825
|
declare const DEFAULT_VOICE_DESIGN_MODEL: VoiceDesignModel;
|
|
826
|
-
/**
|
|
826
|
+
/**
|
|
827
|
+
* I2I providers that take an inpainting mask (`maskUrl`, white = the area to
|
|
828
|
+
* change, black = keep) as the model's own mask. The provider converts the
|
|
829
|
+
* mask into the form its model reads. Every catalog entry listing the
|
|
830
|
+
* `"mask"` feature (`/v1/models`, MCP `list_models`) is one of these.
|
|
831
|
+
*/
|
|
827
832
|
declare const I2I_MASK_SUPPORT: Set<string>;
|
|
828
833
|
/**
|
|
829
834
|
* Mask edit tier per image-gen provider (single source of truth for inpaint).
|
package/package.json
CHANGED
|
@@ -208,6 +208,13 @@ describe("GPT Image 2.6: the images a run sends", () => {
|
|
|
208
208
|
expect(gptImage26EstimatedImages(edit, { referenceImageUrl: A }, 0)).toBe(2)
|
|
209
209
|
})
|
|
210
210
|
|
|
211
|
+
it("a mask is not an input image: neither a painted one nor a wired one is counted", () => {
|
|
212
|
+
const edit = "gpt-image-2-6-flare-i2i"
|
|
213
|
+
const wires = [{ target: "n", targetHandle: "image" }, { target: "n", targetHandle: "mask" }]
|
|
214
|
+
expect(countImageInputEdges(wires, "n")).toBe(0)
|
|
215
|
+
expect(gptImage26EstimatedImages(edit, { maskUrl: A }, countImageInputEdges(wires, "n"))).toBe(1)
|
|
216
|
+
})
|
|
217
|
+
|
|
211
218
|
it("wires count reference images, identity nodes and every image wire after the source, for this node only", () => {
|
|
212
219
|
const edges = [
|
|
213
220
|
{ target: "n", targetHandle: "references" },
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import { describe, it, expect } from "vitest"
|
|
2
|
-
import { IMAGE_GEN_PROVIDERS } from "../model-constants.js"
|
|
2
|
+
import { I2I_MASK_SUPPORT, IMAGE_GEN_PROVIDERS, MODIFY_IMAGE_PROVIDERS } from "../model-constants.js"
|
|
3
3
|
import { IMAGE_MASK_MODE } from "../model-constants.js"
|
|
4
|
+
import { MODEL_CATALOG } from "../model-catalog.js"
|
|
4
5
|
|
|
5
6
|
describe("IMAGE_MASK_MODE", () => {
|
|
6
7
|
it("has an entry for every image-gen provider (completeness invariant)", () => {
|
|
@@ -14,3 +15,35 @@ describe("IMAGE_MASK_MODE", () => {
|
|
|
14
15
|
for (const v of values) expect(["prompt", "composite"]).toContain(v)
|
|
15
16
|
})
|
|
16
17
|
})
|
|
18
|
+
|
|
19
|
+
describe("I2I_MASK_SUPPORT (the image-to-image models that take a mask of their own)", () => {
|
|
20
|
+
it("lists GPT Image 2.6 Flare and Sunburst edits beside Ideogram Edit and FLUX Fill", () => {
|
|
21
|
+
expect([...I2I_MASK_SUPPORT].sort()).toEqual([
|
|
22
|
+
"flux-fill",
|
|
23
|
+
"gpt-image-2-6-flare-i2i",
|
|
24
|
+
"gpt-image-2-6-sunburst-i2i",
|
|
25
|
+
"ideogram-edit",
|
|
26
|
+
])
|
|
27
|
+
})
|
|
28
|
+
|
|
29
|
+
it("leaves out the GPT Image edits that take no mask", () => {
|
|
30
|
+
for (const id of ["gpt-image-2-6-economy-i2i", "gpt-image-2-5-flare-i2i", "gpt-image-2-5-sunburst-i2i", "gpt-image-2-i2i"]) {
|
|
31
|
+
expect(I2I_MASK_SUPPORT.has(id)).toBe(false)
|
|
32
|
+
}
|
|
33
|
+
})
|
|
34
|
+
|
|
35
|
+
it("names only models /v1/image-to-image takes", () => {
|
|
36
|
+
expect([...I2I_MASK_SUPPORT].filter((id) => !(MODIFY_IMAGE_PROVIDERS as readonly string[]).includes(id))).toEqual([])
|
|
37
|
+
})
|
|
38
|
+
|
|
39
|
+
// One way only: FLUX Fill takes a mask but has no catalog entry.
|
|
40
|
+
it("holds every catalog entry that lists the mask feature", () => {
|
|
41
|
+
const listed = Object.values(MODEL_CATALOG).filter((m) => m.features?.includes("mask")).map((m) => m.id)
|
|
42
|
+
expect(listed.filter((id) => !I2I_MASK_SUPPORT.has(id))).toEqual([])
|
|
43
|
+
})
|
|
44
|
+
|
|
45
|
+
it("is listed with the mask feature by every catalog entry it names", () => {
|
|
46
|
+
const inCatalog = [...I2I_MASK_SUPPORT].filter((id) => MODEL_CATALOG[id] !== undefined)
|
|
47
|
+
expect(inCatalog.filter((id) => !MODEL_CATALOG[id]!.features?.includes("mask"))).toEqual([])
|
|
48
|
+
})
|
|
49
|
+
})
|
|
@@ -0,0 +1,150 @@
|
|
|
1
|
+
import { describe, expect, it } from "vitest"
|
|
2
|
+
import { MODEL_CATALOG, maskEditMatchesSource, normalizeModelInput, resolutionNearestSource, sourceImageSizeNeeded } from "../model-catalog.js"
|
|
3
|
+
import { resolveNormalizedImageGen } from "../credit-identifiers.js"
|
|
4
|
+
import { I2I_MASK_SUPPORT } from "../model-constants.js"
|
|
5
|
+
import { nodeIsMasked, wiredImageSize } from "../wired-image-size.js"
|
|
6
|
+
|
|
7
|
+
/**
|
|
8
|
+
* A masked GPT Image 2.6 edit follows its photo: the result is pasted back onto
|
|
9
|
+
* the photo at the photo's own size, so the run keeps the photo's shape and is
|
|
10
|
+
* rendered, and priced, at the size nearest the photo.
|
|
11
|
+
*/
|
|
12
|
+
const FOLLOWING = ["gpt-image-2-6-flare-i2i", "gpt-image-2-6-sunburst-i2i"] as const
|
|
13
|
+
const PHOTO_1K_PORTRAIT = { width: 768, height: 1024 }
|
|
14
|
+
const PHOTO_4000_PORTRAIT = { width: 3000, height: 4000 }
|
|
15
|
+
|
|
16
|
+
describe("the catalog", () => {
|
|
17
|
+
it("only GPT Image 2.6 Flare's and Sunburst's edits follow their photo", () => {
|
|
18
|
+
const declared = Object.values(MODEL_CATALOG).filter((m) => m.maskEditMatchesSource).map((m) => m.id).sort()
|
|
19
|
+
expect(declared).toEqual([...FOLLOWING])
|
|
20
|
+
})
|
|
21
|
+
|
|
22
|
+
it("each such model takes a mask, and has a size for every ratio it lists at every size", () => {
|
|
23
|
+
for (const id of FOLLOWING) {
|
|
24
|
+
const m = MODEL_CATALOG[id]!
|
|
25
|
+
expect(I2I_MASK_SUPPORT.has(id)).toBe(true)
|
|
26
|
+
for (const ratio of m.aspectRatios!.filter((r) => r !== "auto")) {
|
|
27
|
+
for (const size of m.resolutions!) expect(m.outputSizes?.[ratio]?.[size], `${id} ${ratio} ${size}`).toBeDefined()
|
|
28
|
+
}
|
|
29
|
+
}
|
|
30
|
+
})
|
|
31
|
+
|
|
32
|
+
it("the table is the provider's: 1:1 at 4K is 2880 square, 9:16 at 4K 2160x3840", () => {
|
|
33
|
+
const sizes = MODEL_CATALOG["gpt-image-2-6-flare-i2i"]!.outputSizes!
|
|
34
|
+
expect(sizes["1:1"]!["4K"]).toEqual([2880, 2880])
|
|
35
|
+
expect(sizes["9:16"]!["4K"]).toEqual([2160, 3840])
|
|
36
|
+
expect(sizes["3:4"]!["1K"]).toEqual([768, 1024])
|
|
37
|
+
})
|
|
38
|
+
})
|
|
39
|
+
|
|
40
|
+
describe("maskEditMatchesSource / sourceImageSizeNeeded", () => {
|
|
41
|
+
it("is true only with a mask, on a model that follows its photo", () => {
|
|
42
|
+
expect(maskEditMatchesSource("gpt-image-2-6-flare-i2i", true)).toBe(true)
|
|
43
|
+
expect(maskEditMatchesSource("gpt-image-2-6-flare-i2i", false)).toBe(false)
|
|
44
|
+
expect(maskEditMatchesSource("ideogram-edit", true)).toBe(false)
|
|
45
|
+
expect(maskEditMatchesSource("gpt-image-2-6-economy-i2i", true)).toBe(false)
|
|
46
|
+
})
|
|
47
|
+
|
|
48
|
+
it("a masked 2.6 edit needs the photo's size even when it names a ratio", () => {
|
|
49
|
+
expect(sourceImageSizeNeeded("gpt-image-2-6-flare-i2i", "16:9", true)).toBe(true)
|
|
50
|
+
expect(sourceImageSizeNeeded("gpt-image-2-6-flare-i2i", "16:9", false)).toBe(false)
|
|
51
|
+
expect(sourceImageSizeNeeded("gpt-image-2-6-flare-i2i", "auto", false)).toBe(true)
|
|
52
|
+
})
|
|
53
|
+
})
|
|
54
|
+
|
|
55
|
+
describe("resolutionNearestSource", () => {
|
|
56
|
+
it("picks the size whose long side is nearest the photo's", () => {
|
|
57
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "3:4", PHOTO_1K_PORTRAIT)).toBe("1K")
|
|
58
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "3:4", PHOTO_4000_PORTRAIT)).toBe("4K")
|
|
59
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "3:4", { width: 1536, height: 2048 })).toBe("2K")
|
|
60
|
+
// Nearest by ratio: 1500 px is nearer 2048 than 1024.
|
|
61
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "1:1", { width: 1500, height: 1500 })).toBe("2K")
|
|
62
|
+
})
|
|
63
|
+
|
|
64
|
+
it("is undefined with no size, an unknown ratio or a model without a table", () => {
|
|
65
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "3:4", undefined)).toBeUndefined()
|
|
66
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "auto", PHOTO_1K_PORTRAIT)).toBeUndefined()
|
|
67
|
+
expect(resolutionNearestSource("gpt-image-2-6-flare-i2i", "constructor", PHOTO_1K_PORTRAIT)).toBeUndefined()
|
|
68
|
+
expect(resolutionNearestSource("gpt-image-2-i2i", "3:4", PHOTO_1K_PORTRAIT)).toBeUndefined()
|
|
69
|
+
})
|
|
70
|
+
})
|
|
71
|
+
|
|
72
|
+
describe("normalizeModelInput: a masked edit that follows its photo", () => {
|
|
73
|
+
for (const id of FOLLOWING) {
|
|
74
|
+
it(`${id}: a 1K photo with a 4K request runs, and is priced, at 1K`, () => {
|
|
75
|
+
const n = normalizeModelInput(id, { aspectRatio: "auto", resolution: "4K" }, { sourceImage: PHOTO_1K_PORTRAIT, masked: true })
|
|
76
|
+
expect(n.aspectRatio).toBe("3:4")
|
|
77
|
+
expect(n.resolution).toBe("1K")
|
|
78
|
+
expect(n.adjustments).toContainEqual(expect.objectContaining({ field: "resolution", from: "4K", to: "1K" }))
|
|
79
|
+
const priced = resolveNormalizedImageGen({ provider: id, aspectRatio: "auto", resolution: "4K", refCount: 1, sourceImage: PHOTO_1K_PORTRAIT, masked: true })
|
|
80
|
+
expect(priced.identifier).toBe(`${id}:medium:1K:1ref`)
|
|
81
|
+
expect(priced.resolution).toBe("1K")
|
|
82
|
+
})
|
|
83
|
+
|
|
84
|
+
it(`${id}: a 4000 px photo with a 1K request runs, and is priced, at 4K`, () => {
|
|
85
|
+
const priced = resolveNormalizedImageGen({ provider: id, aspectRatio: "auto", resolution: "1K", refCount: 1, sourceImage: PHOTO_4000_PORTRAIT, masked: true })
|
|
86
|
+
expect(priced.identifier).toBe(`${id}:medium:4K:1ref`)
|
|
87
|
+
expect(priced.adjustments).toContainEqual(expect.objectContaining({ field: "resolution", from: "1K", to: "4K" }))
|
|
88
|
+
})
|
|
89
|
+
|
|
90
|
+
it(`${id}: a named 16:9 on a 3:4 photo runs 3:4, and says so`, () => {
|
|
91
|
+
const n = normalizeModelInput(id, { aspectRatio: "16:9", resolution: "1K" }, { sourceImage: PHOTO_1K_PORTRAIT, masked: true })
|
|
92
|
+
expect(n.aspectRatio).toBe("3:4")
|
|
93
|
+
expect(n.adjustments).toContainEqual(expect.objectContaining({ field: "aspectRatio", from: "16:9", to: "3:4" }))
|
|
94
|
+
expect(n.adjustments.find((a) => a.field === "aspectRatio")!.reason).toContain("masked edit")
|
|
95
|
+
})
|
|
96
|
+
|
|
97
|
+
it(`${id}: no ratio named also keeps the photo's shape`, () => {
|
|
98
|
+
const n = normalizeModelInput(id, { resolution: "2K" }, { sourceImage: { width: 2048, height: 1536 }, masked: true })
|
|
99
|
+
expect(n.aspectRatio).toBe("4:3")
|
|
100
|
+
expect(n.resolution).toBe("2K")
|
|
101
|
+
expect(n.adjustments).toEqual([])
|
|
102
|
+
})
|
|
103
|
+
|
|
104
|
+
it(`${id}: with the photo's size unknown, the ratio is "auto" and the size asked is kept`, () => {
|
|
105
|
+
const n = normalizeModelInput(id, { aspectRatio: "16:9", resolution: "4K" }, { masked: true })
|
|
106
|
+
expect(n.aspectRatio).toBe("auto")
|
|
107
|
+
expect(n.resolution).toBe("4K")
|
|
108
|
+
expect(n.adjustments).toEqual([expect.objectContaining({ field: "aspectRatio", from: "16:9", to: "auto" })])
|
|
109
|
+
})
|
|
110
|
+
|
|
111
|
+
it(`${id}: without a mask, the ratio and the size asked are kept`, () => {
|
|
112
|
+
const n = normalizeModelInput(id, { aspectRatio: "16:9", resolution: "4K" }, { sourceImage: PHOTO_1K_PORTRAIT })
|
|
113
|
+
expect(n.aspectRatio).toBe("16:9")
|
|
114
|
+
expect(n.resolution).toBe("4K")
|
|
115
|
+
expect(n.adjustments).toEqual([])
|
|
116
|
+
})
|
|
117
|
+
|
|
118
|
+
it(`${id}: a transparent background still renders at 1K`, () => {
|
|
119
|
+
const n = normalizeModelInput(id, { aspectRatio: "auto", resolution: "1K", background: "transparent" }, { sourceImage: PHOTO_4000_PORTRAIT, masked: true })
|
|
120
|
+
expect(n.resolution).toBe("1K")
|
|
121
|
+
})
|
|
122
|
+
}
|
|
123
|
+
|
|
124
|
+
it("another mask model keeps the ratio and the size it names", () => {
|
|
125
|
+
const n = normalizeModelInput("gpt-image-2-6-economy-i2i", { aspectRatio: "16:9" }, { sourceImage: PHOTO_1K_PORTRAIT, masked: true })
|
|
126
|
+
expect(n.aspectRatio).toBe("16:9")
|
|
127
|
+
})
|
|
128
|
+
})
|
|
129
|
+
|
|
130
|
+
describe("wiredImageSize / nodeIsMasked (what an estimate knows of the photo)", () => {
|
|
131
|
+
const nodes = [
|
|
132
|
+
{ id: "gen", data: { generatedResults: [{ width: 1024, height: 1024 }, { width: 768, height: 1024 }], activeResultIndex: 1 } },
|
|
133
|
+
{ id: "up", data: { width: 4000, height: 3000 } },
|
|
134
|
+
{ id: "none", data: {} },
|
|
135
|
+
{ id: "n", data: {} },
|
|
136
|
+
]
|
|
137
|
+
it("reads the shown result's size, else the node's own, from the node wired into `image`", () => {
|
|
138
|
+
expect(wiredImageSize("n", nodes, [{ source: "gen", target: "n", targetHandle: "image" }])).toEqual({ width: 768, height: 1024 })
|
|
139
|
+
expect(wiredImageSize("n", nodes, [{ source: "up", target: "n", targetHandle: "image" }])).toEqual({ width: 4000, height: 3000 })
|
|
140
|
+
expect(wiredImageSize("n", nodes, [{ source: "none", target: "n", targetHandle: "image" }])).toBeUndefined()
|
|
141
|
+
// A mask or another input is not the photo.
|
|
142
|
+
expect(wiredImageSize("n", nodes, [{ source: "up", target: "n", targetHandle: "mask" }])).toBeUndefined()
|
|
143
|
+
})
|
|
144
|
+
|
|
145
|
+
it("a node is masked by its own maskUrl or a wire into `mask`", () => {
|
|
146
|
+
expect(nodeIsMasked("n", { maskUrl: "https://m.png" }, [])).toBe(true)
|
|
147
|
+
expect(nodeIsMasked("n", {}, [{ source: "gm", target: "n", targetHandle: "mask" }])).toBe(true)
|
|
148
|
+
expect(nodeIsMasked("n", {}, [{ source: "x", target: "n", targetHandle: "image" }])).toBe(false)
|
|
149
|
+
})
|
|
150
|
+
})
|
|
@@ -289,10 +289,12 @@ export function imageFormatInputOf(source: object): { outputFormat?: string; gif
|
|
|
289
289
|
*
|
|
290
290
|
* `sourceImage` is the size of the photo the request transforms or is drawn
|
|
291
291
|
* from (an image-to-image / edit source, or Generate Image's first wired
|
|
292
|
-
* reference). It
|
|
292
|
+
* reference). It decides "auto" on a model without a native auto, and the
|
|
293
293
|
* ratio is not a pricing dimension on any such model (pinned across the catalog
|
|
294
|
-
* by `auto-aspect-source.test.ts`)
|
|
295
|
-
*
|
|
294
|
+
* by `auto-aspect-source.test.ts`). With `masked` on a model whose masked edit
|
|
295
|
+
* follows its photo (`maskEditMatchesSource`) it also decides the SIZE, a
|
|
296
|
+
* pricing dimension: the CHECK and the DEBIT must then both be given it (the
|
|
297
|
+
* image-to-image route reads it once, before its credit check).
|
|
296
298
|
*/
|
|
297
299
|
export function resolveNormalizedImageGen(opts: {
|
|
298
300
|
provider: string | undefined
|
|
@@ -310,6 +312,8 @@ export function resolveNormalizedImageGen(opts: {
|
|
|
310
312
|
refCount: number
|
|
311
313
|
swapToI2i?: boolean
|
|
312
314
|
sourceImage?: SourceImageSize
|
|
315
|
+
/** The request carries an inpainting mask: on a model whose masked edit follows its photo, the photo's size decides the ratio and the priced size. */
|
|
316
|
+
masked?: boolean
|
|
313
317
|
}): NormalizedImageGen {
|
|
314
318
|
const str = (v: unknown): string | undefined =>
|
|
315
319
|
typeof v === "string" && v.length > 0 ? v : undefined
|
|
@@ -325,7 +329,7 @@ export function resolveNormalizedImageGen(opts: {
|
|
|
325
329
|
background: str(opts.background),
|
|
326
330
|
...imageFormatInputOf(opts),
|
|
327
331
|
},
|
|
328
|
-
{ sourceImage: opts.sourceImage },
|
|
332
|
+
{ sourceImage: opts.sourceImage, masked: opts.masked },
|
|
329
333
|
)
|
|
330
334
|
|
|
331
335
|
return {
|
|
@@ -78,6 +78,10 @@ export {
|
|
|
78
78
|
defaultResolutionFor,
|
|
79
79
|
nearestCatalogAspectRatio,
|
|
80
80
|
autoAspectNeedsSourceImage,
|
|
81
|
+
maskEditMatchesSource,
|
|
82
|
+
sourceImageSizeNeeded,
|
|
83
|
+
resolutionNearestSource,
|
|
84
|
+
maskedEditResolution,
|
|
81
85
|
isAutoAspectToken,
|
|
82
86
|
resolutionsAtAspectRatio,
|
|
83
87
|
resolutionAtAspectRatio,
|
package/src/entries/workflow.ts
CHANGED
|
@@ -145,6 +145,7 @@ export { resolveFieldMappings, resolveLocationFields, type FieldEdgeSource } fro
|
|
|
145
145
|
export { resolveNodeRefs, parseNodeRef, canonicalVarName, NODE_REF_PATTERN, RESERVED_TEMPLATE_VARS, extractReferencedLabels, combineSameLabelRefs, refHandleCategory, REF_HANDLE_CATEGORY, REFERENCE_HANDLE_MAP, referenceModalityForHandle, FRAME_TARGET_HANDLES, countImageInputEdges, countRefModalityEdges, REF_TOKEN_NAMESPACE_PREFIXES, classifyRefToken, unresolvedRefTokens } from "../node-refs.js"
|
|
146
146
|
export type { RefCandidate, ReferenceModality, RefModalityEdge, RefTokenKind } from "../node-refs.js"
|
|
147
147
|
export { resolveSourceThroughConnectedList } from "../list-source-resolver.js"
|
|
148
|
+
export { wiredImageSize, recordedImageSize, nodeIsMasked, type SizedGraphNode, type SizedGraphEdge } from "../wired-image-size.js"
|
|
148
149
|
export { zipMergeLists } from "../list-merge.js"
|
|
149
150
|
export type {
|
|
150
151
|
WorkflowExport,
|
package/src/model-catalog.ts
CHANGED
|
@@ -327,6 +327,21 @@ export interface ModelCatalogEntry {
|
|
|
327
327
|
* as before. "auto" stays in `aspectRatios`, so every menu still offers it.
|
|
328
328
|
*/
|
|
329
329
|
autoAspectFromSourceImage?: true
|
|
330
|
+
/**
|
|
331
|
+
* The pixel size (width, height) this model renders at each aspect ratio and
|
|
332
|
+
* size, from the provider's own table, for a PNG, JPEG or WebP. Read where a
|
|
333
|
+
* run's size is chosen from its photo's size (`maskEditMatchesSource`).
|
|
334
|
+
*/
|
|
335
|
+
outputSizes?: Readonly<Record<string, Readonly<Record<string, readonly [number, number]>>>>
|
|
336
|
+
/**
|
|
337
|
+
* A masked edit on this model follows its photo: the result is pasted back
|
|
338
|
+
* onto the photo at the photo's own size, so the run keeps the photo's shape
|
|
339
|
+
* ("auto", whatever ratio it names) and renders at the size of `outputSizes`
|
|
340
|
+
* whose long side is nearest the photo's, priced at that size
|
|
341
|
+
* (`normalizeModelInput` with `context.masked`; `resolutionNearestSource`).
|
|
342
|
+
* With the photo's size unknown, the size asked is kept.
|
|
343
|
+
*/
|
|
344
|
+
maskEditMatchesSource?: true
|
|
330
345
|
resolutions?: readonly string[]
|
|
331
346
|
/**
|
|
332
347
|
* The aspect ratios at which this model renders fewer sizes than
|
|
@@ -550,6 +565,26 @@ const GPT_IMAGE_2_6_RATIOS = [
|
|
|
550
565
|
"2:1", "1:2", "21:9", "9:21", "3:1", "1:3",
|
|
551
566
|
] as const
|
|
552
567
|
const GPT_IMAGE_2_6_ECONOMY_RATIOS = ["auto", "1:1", "16:9", "9:16", "3:2", "2:3", "21:9", "4:3"] as const
|
|
568
|
+
// What GPT Image 2.6 Flare and Sunburst render at each ratio and size, for a
|
|
569
|
+
// PNG, JPEG or WebP (the `resolution` table of docs.kie.ai/market/gpt/
|
|
570
|
+
// gpt-image-2-6-{flare,sunburst}-image-to-image, fetched 2026-11-11).
|
|
571
|
+
const GPT_IMAGE_2_6_OUTPUT_SIZES = {
|
|
572
|
+
"1:1": { "1K": [1024, 1024], "2K": [2048, 2048], "4K": [2880, 2880] },
|
|
573
|
+
"3:2": { "1K": [1536, 1024], "2K": [2048, 1360], "4K": [3520, 2336] },
|
|
574
|
+
"2:3": { "1K": [1024, 1536], "2K": [1360, 2048], "4K": [2336, 3520] },
|
|
575
|
+
"4:3": { "1K": [1024, 768], "2K": [2048, 1536], "4K": [3312, 2480] },
|
|
576
|
+
"3:4": { "1K": [768, 1024], "2K": [1536, 2048], "4K": [2480, 3312] },
|
|
577
|
+
"5:4": { "1K": [1280, 1024], "2K": [2560, 2048], "4K": [3216, 2576] },
|
|
578
|
+
"4:5": { "1K": [1024, 1280], "2K": [2048, 2560], "4K": [2576, 3216] },
|
|
579
|
+
"16:9": { "1K": [1536, 864], "2K": [2048, 1152], "4K": [3840, 2160] },
|
|
580
|
+
"9:16": { "1K": [864, 1536], "2K": [1152, 2048], "4K": [2160, 3840] },
|
|
581
|
+
"2:1": { "1K": [2048, 1024], "2K": [2688, 1344], "4K": [3840, 1920] },
|
|
582
|
+
"1:2": { "1K": [1024, 2048], "2K": [1344, 2688], "4K": [1920, 3840] },
|
|
583
|
+
"21:9": { "1K": [2016, 864], "2K": [2688, 1152], "4K": [3840, 1648] },
|
|
584
|
+
"9:21": { "1K": [864, 2016], "2K": [1152, 2688], "4K": [1648, 3840] },
|
|
585
|
+
"3:1": { "1K": [1536, 512], "2K": [3072, 1024], "4K": [3840, 1280] },
|
|
586
|
+
"1:3": { "1K": [512, 1536], "2K": [1024, 3072], "4K": [1280, 3840] },
|
|
587
|
+
} as const satisfies Record<string, Record<string, readonly [number, number]>>
|
|
553
588
|
const GPT_IMAGE_2_6_VALUE_LABELS = {
|
|
554
589
|
"3:2": "3:2 (Landscape)", "2:3": "2:3 (Portrait)",
|
|
555
590
|
"5:4": "5:4 (Near-square)", "4:5": "4:5 (Near-square)",
|
|
@@ -1175,11 +1210,13 @@ const IMAGE_MODELS: Record<string, ModelCatalogEntry> = {
|
|
|
1175
1210
|
maker: "openai",
|
|
1176
1211
|
label: "GPT Image 2.6 Flare (I2I)",
|
|
1177
1212
|
series: "GPT Image",
|
|
1178
|
-
description: "Fast GPT Image 2.6 edits - up to 16 input images, priced per image. Writes PNG, JPEG, WebP, an animated GIF or an SVG. Pick it while iterating.",
|
|
1179
|
-
useCases: ["edit", "high-res"],
|
|
1180
|
-
features: ["reference-image"],
|
|
1213
|
+
description: "Fast GPT Image 2.6 edits - up to 16 input images, priced per image. Writes PNG, JPEG, WebP, an animated GIF or an SVG. Takes an inpainting mask (PNG, JPEG or WebP; the rest of the picture is kept as it was). Pick it while iterating.",
|
|
1214
|
+
useCases: ["edit", "inpaint", "high-res"],
|
|
1215
|
+
features: ["reference-image", "mask"],
|
|
1181
1216
|
aspectRatios: GPT_IMAGE_2_6_RATIOS,
|
|
1182
1217
|
autoAspectFromSourceImage: true,
|
|
1218
|
+
outputSizes: GPT_IMAGE_2_6_OUTPUT_SIZES,
|
|
1219
|
+
maskEditMatchesSource: true,
|
|
1183
1220
|
resolutions: ["1K", "2K", "4K"],
|
|
1184
1221
|
qualities: ["auto", "low", "medium", "high", "xhigh", "max"],
|
|
1185
1222
|
defaultQuality: "medium",
|
|
@@ -1283,11 +1320,13 @@ const IMAGE_MODELS: Record<string, ModelCatalogEntry> = {
|
|
|
1283
1320
|
maker: "openai",
|
|
1284
1321
|
label: "GPT Image 2.6 Sunburst (I2I)",
|
|
1285
1322
|
series: "GPT Image",
|
|
1286
|
-
description: "Precision GPT Image 2.6 edits - up to 16 input images, priced per image. Writes PNG, JPEG, WebP, an animated GIF or an SVG. Pick it for finished, detail-critical edits.",
|
|
1287
|
-
useCases: ["edit", "high-res"],
|
|
1288
|
-
features: ["reference-image"],
|
|
1323
|
+
description: "Precision GPT Image 2.6 edits - up to 16 input images, priced per image. Writes PNG, JPEG, WebP, an animated GIF or an SVG. Takes an inpainting mask (PNG, JPEG or WebP; the rest of the picture is kept as it was). Pick it for finished, detail-critical edits.",
|
|
1324
|
+
useCases: ["edit", "inpaint", "high-res"],
|
|
1325
|
+
features: ["reference-image", "mask"],
|
|
1289
1326
|
aspectRatios: GPT_IMAGE_2_6_RATIOS,
|
|
1290
1327
|
autoAspectFromSourceImage: true,
|
|
1328
|
+
outputSizes: GPT_IMAGE_2_6_OUTPUT_SIZES,
|
|
1329
|
+
maskEditMatchesSource: true,
|
|
1291
1330
|
resolutions: ["1K", "2K", "4K"],
|
|
1292
1331
|
qualities: ["auto", "low", "medium", "high", "xhigh", "max"],
|
|
1293
1332
|
defaultQuality: "medium",
|
|
@@ -4183,8 +4222,80 @@ export interface ModelInputContext {
|
|
|
4183
4222
|
* resolve "auto" on a model without a native auto, or whose own auto does
|
|
4184
4223
|
* not keep the photo's shape (`autoAspectFromSourceImage`); absent or
|
|
4185
4224
|
* unusable means the size is unknown and "auto" snaps the way it always has.
|
|
4225
|
+
* Also read by a masked edit on a model whose result follows its photo
|
|
4226
|
+
* (`maskEditMatchesSource`): it decides the ratio and the size.
|
|
4186
4227
|
*/
|
|
4187
4228
|
sourceImage?: SourceImageSize
|
|
4229
|
+
/** The request carries an inpainting mask (`maskUrl`). */
|
|
4230
|
+
masked?: boolean
|
|
4231
|
+
}
|
|
4232
|
+
|
|
4233
|
+
/**
|
|
4234
|
+
* True when a masked request on `modelId` follows its photo
|
|
4235
|
+
* (`maskEditMatchesSource`): its ratio and size come from the photo's size.
|
|
4236
|
+
*/
|
|
4237
|
+
export function maskEditMatchesSource(modelId: string | undefined, masked: unknown): boolean {
|
|
4238
|
+
return masked === true && !!modelId && MODEL_CATALOG[modelId]?.maskEditMatchesSource === true
|
|
4239
|
+
}
|
|
4240
|
+
|
|
4241
|
+
/**
|
|
4242
|
+
* True when the photo's size can change what a request on `modelId` sends:
|
|
4243
|
+
* "auto" that keeps the photo's shape (`autoAspectNeedsSourceImage`), or a
|
|
4244
|
+
* masked edit that follows its photo (`maskEditMatchesSource`). A caller that
|
|
4245
|
+
* has to fetch the size asks this first.
|
|
4246
|
+
*/
|
|
4247
|
+
export function sourceImageSizeNeeded(modelId: string | undefined, aspectRatio: unknown, masked: unknown): boolean {
|
|
4248
|
+
return autoAspectNeedsSourceImage(modelId, aspectRatio) || maskEditMatchesSource(modelId, masked)
|
|
4249
|
+
}
|
|
4250
|
+
|
|
4251
|
+
/**
|
|
4252
|
+
* The size a MASKED request on `modelId` runs, and is priced, at once its
|
|
4253
|
+
* photo's size is known — what `normalizeModelInput` sends for it. For an
|
|
4254
|
+
* estimate that knows the photo's size (the editor's badge, a workflow
|
|
4255
|
+
* estimate). `undefined` when the model does not follow its photo
|
|
4256
|
+
* (`maskEditMatchesSource`) or the size is unusable: the caller keeps its own.
|
|
4257
|
+
*/
|
|
4258
|
+
export function maskedEditResolution(
|
|
4259
|
+
modelId: string,
|
|
4260
|
+
levers: { aspectRatio?: unknown; resolution?: unknown; background?: unknown; outputFormat?: unknown },
|
|
4261
|
+
sourceImage: SourceImageSize | undefined,
|
|
4262
|
+
): string | undefined {
|
|
4263
|
+
if (!maskEditMatchesSource(modelId, true) || !usableSize(sourceImage)) return undefined
|
|
4264
|
+
const str = (v: unknown): string | undefined => (typeof v === "string" && v.length > 0 ? v : undefined)
|
|
4265
|
+
return normalizeModelInput(
|
|
4266
|
+
modelId,
|
|
4267
|
+
{ aspectRatio: str(levers.aspectRatio), resolution: str(levers.resolution), background: str(levers.background), outputFormat: str(levers.outputFormat) },
|
|
4268
|
+
{ sourceImage, masked: true },
|
|
4269
|
+
).resolution
|
|
4270
|
+
}
|
|
4271
|
+
|
|
4272
|
+
/**
|
|
4273
|
+
* The size of `modelId` whose rendered long side, at `aspectRatio`, is nearest
|
|
4274
|
+
* the photo's long side (`outputSizes`; nearest by ratio, so 1500 px is nearer
|
|
4275
|
+
* 2048 than 1024; a tie goes to the larger size). `undefined` when the model
|
|
4276
|
+
* has no size table for that ratio or the photo's size is unusable.
|
|
4277
|
+
*/
|
|
4278
|
+
export function resolutionNearestSource(
|
|
4279
|
+
modelId: string,
|
|
4280
|
+
aspectRatio: unknown,
|
|
4281
|
+
source: SourceImageSize | undefined,
|
|
4282
|
+
): string | undefined {
|
|
4283
|
+
const m = MODEL_CATALOG[modelId]
|
|
4284
|
+
const table = m?.outputSizes
|
|
4285
|
+
if (!table || typeof aspectRatio !== "string" || !usableSize(source)) return undefined
|
|
4286
|
+
// Own keys only: a ratio such as "constructor" must not read the prototype.
|
|
4287
|
+
if (!Object.prototype.hasOwnProperty.call(table, aspectRatio)) return undefined
|
|
4288
|
+
const sizes = table[aspectRatio]!
|
|
4289
|
+
const long = Math.max(source.width, source.height)
|
|
4290
|
+
let best: string | undefined
|
|
4291
|
+
let bestDist = Infinity
|
|
4292
|
+
for (const resolution of m.resolutions ?? Object.keys(sizes)) {
|
|
4293
|
+
const px = Object.prototype.hasOwnProperty.call(sizes, resolution) ? sizes[resolution] : undefined
|
|
4294
|
+
if (!px) continue
|
|
4295
|
+
const dist = Math.abs(Math.log(Math.max(px[0], px[1]) / long))
|
|
4296
|
+
if (dist <= bestDist + ASPECT_TIE_EPSILON) { bestDist = dist; best = resolution }
|
|
4297
|
+
}
|
|
4298
|
+
return best
|
|
4188
4299
|
}
|
|
4189
4300
|
|
|
4190
4301
|
/** True for the "auto" aspect token in any case or padding. */
|
|
@@ -4392,12 +4503,31 @@ export function normalizeModelInput(
|
|
|
4392
4503
|
out.gifFrameDurationMs = format.levers.gifFrameDurationMs
|
|
4393
4504
|
out.svgLevel = format.levers.svgLevel
|
|
4394
4505
|
const gif = format.levers.outputFormat === "gif"
|
|
4506
|
+
// A masked edit that follows its photo runs at "auto" whatever ratio it
|
|
4507
|
+
// names, and at the size nearest the photo (`maskEditMatchesSource`).
|
|
4508
|
+
const followsSource = !gif && maskEditMatchesSource(modelId, context.masked)
|
|
4509
|
+
const askedRatio = followsSource ? "auto" : input.aspectRatio
|
|
4395
4510
|
|
|
4396
4511
|
const sourceRatio =
|
|
4397
|
-
!gif && autoAspectNeedsSourceImage(modelId,
|
|
4512
|
+
!gif && autoAspectNeedsSourceImage(modelId, askedRatio) && usableSize(context.sourceImage)
|
|
4398
4513
|
? nearestCatalogAspectRatio(context.sourceImage.width, context.sourceImage.height, m.aspectRatios ?? [])
|
|
4399
4514
|
: undefined
|
|
4400
|
-
if (
|
|
4515
|
+
if (followsSource && input.aspectRatio !== undefined && !isAutoAspectToken(input.aspectRatio)) {
|
|
4516
|
+
const to = sourceRatio ?? "auto"
|
|
4517
|
+
if (to !== input.aspectRatio) {
|
|
4518
|
+
adjustments.push({
|
|
4519
|
+
field: "aspectRatio",
|
|
4520
|
+
from: input.aspectRatio,
|
|
4521
|
+
to,
|
|
4522
|
+
reason: sourceRatio !== undefined
|
|
4523
|
+
? `A masked edit on ${m.label} keeps the photo's shape — using "${sourceRatio}", the supported ratio closest to the source image (${context.sourceImage!.width}x${context.sourceImage!.height}).`
|
|
4524
|
+
: `A masked edit on ${m.label} keeps the photo's shape — using "auto".`,
|
|
4525
|
+
})
|
|
4526
|
+
}
|
|
4527
|
+
out.aspectRatio = to
|
|
4528
|
+
} else if (followsSource && input.aspectRatio === undefined) {
|
|
4529
|
+
out.aspectRatio = sourceRatio ?? "auto"
|
|
4530
|
+
} else if (gif) {
|
|
4401
4531
|
// A GIF is square, IMAGE_GIF_FRAME_SIZE a frame: neither lever applies.
|
|
4402
4532
|
// "auto" leaves the shape to the model, and square is its answer.
|
|
4403
4533
|
if (input.aspectRatio !== undefined && input.aspectRatio !== "1:1" && !isAutoAspectToken(input.aspectRatio)) {
|
|
@@ -4409,14 +4539,14 @@ export function normalizeModelInput(
|
|
|
4409
4539
|
})
|
|
4410
4540
|
}
|
|
4411
4541
|
out.aspectRatio = input.aspectRatio === undefined ? undefined : "1:1"
|
|
4412
|
-
} else if (sourceRatio !== undefined &&
|
|
4542
|
+
} else if (sourceRatio !== undefined && askedRatio !== undefined) {
|
|
4413
4543
|
const { width, height } = context.sourceImage!
|
|
4414
4544
|
const why = m.autoAspectFromSourceImage === true
|
|
4415
4545
|
? `${m.label}'s own "auto" does not keep the photo's shape`
|
|
4416
4546
|
: `${m.label} has no "auto" aspect ratio`
|
|
4417
4547
|
adjustments.push({
|
|
4418
4548
|
field: "aspectRatio",
|
|
4419
|
-
from: input.aspectRatio,
|
|
4549
|
+
from: input.aspectRatio ?? askedRatio,
|
|
4420
4550
|
to: sourceRatio,
|
|
4421
4551
|
reason: `${why} — using "${sourceRatio}", the supported ratio closest to the source image's shape (${width}x${height}).`,
|
|
4422
4552
|
})
|
|
@@ -4424,7 +4554,23 @@ export function normalizeModelInput(
|
|
|
4424
4554
|
} else {
|
|
4425
4555
|
out.aspectRatio = snap("aspectRatio", input.aspectRatio, m.aspectRatios)
|
|
4426
4556
|
}
|
|
4427
|
-
if (
|
|
4557
|
+
if (followsSource) {
|
|
4558
|
+
// The size whose long side is nearest the photo's: the result is the
|
|
4559
|
+
// photo's own size, so the run is rendered, and priced, near it.
|
|
4560
|
+
const asked = snap("resolution", input.resolution, m.resolutions, defaultResolutionFor(modelId))
|
|
4561
|
+
const nearest = resolutionNearestSource(modelId, out.aspectRatio, context.sourceImage)
|
|
4562
|
+
const from = asked ?? m.defaultResolution ?? m.resolutions?.[0]
|
|
4563
|
+
if (nearest !== undefined && from !== undefined && nearest !== from) {
|
|
4564
|
+
const { width, height } = context.sourceImage!
|
|
4565
|
+
adjustments.push({
|
|
4566
|
+
field: "resolution",
|
|
4567
|
+
from,
|
|
4568
|
+
to: nearest,
|
|
4569
|
+
reason: `A masked edit on ${m.label} is delivered at the photo's own size (${width}x${height}) — rendering at ${nearest}, the size nearest it.`,
|
|
4570
|
+
})
|
|
4571
|
+
}
|
|
4572
|
+
out.resolution = nearest ?? asked
|
|
4573
|
+
} else if (gif) {
|
|
4428
4574
|
if (input.resolution !== undefined) {
|
|
4429
4575
|
adjustments.push({
|
|
4430
4576
|
field: "resolution",
|
package/src/model-constants.ts
CHANGED
|
@@ -1820,8 +1820,18 @@ export type VoiceDesignModel = typeof VOICE_DESIGN_MODELS[number]
|
|
|
1820
1820
|
*/
|
|
1821
1821
|
export const DEFAULT_VOICE_DESIGN_MODEL: VoiceDesignModel = "eleven_ttv_v3"
|
|
1822
1822
|
|
|
1823
|
-
/**
|
|
1824
|
-
|
|
1823
|
+
/**
|
|
1824
|
+
* I2I providers that take an inpainting mask (`maskUrl`, white = the area to
|
|
1825
|
+
* change, black = keep) as the model's own mask. The provider converts the
|
|
1826
|
+
* mask into the form its model reads. Every catalog entry listing the
|
|
1827
|
+
* `"mask"` feature (`/v1/models`, MCP `list_models`) is one of these.
|
|
1828
|
+
*/
|
|
1829
|
+
export const I2I_MASK_SUPPORT = new Set([
|
|
1830
|
+
"ideogram-edit",
|
|
1831
|
+
"flux-fill",
|
|
1832
|
+
"gpt-image-2-6-flare-i2i",
|
|
1833
|
+
"gpt-image-2-6-sunburst-i2i",
|
|
1834
|
+
])
|
|
1825
1835
|
|
|
1826
1836
|
/**
|
|
1827
1837
|
* Mask edit tier per image-gen provider (single source of truth for inpaint).
|