@sogni-ai/sogni-client 5.83.0 → 5.84.0

package/src/Projects/types/index.ts

@@ -372,15 +372,20 @@ export interface MinimaxH3Keyframe {
  *   two-stage ID (`minimax-h3-fastvideo-omni-int8_r2v_turbo_2stage`) takes the
  *   same request and delivers twice the canvas, with the same 384/544/768 px
  *   canvas choices as the other Two-Stage IDs.
+ * - Prism (`minimax-h3-prism-int8_r2v`) is an optional `r2v` model for mature
+ *   themes with its recipe fixed at 8 Euler steps. It takes the same
+ *   references, keyframes, catalog LoRAs, frame grid and canvas as the other
+ *   one-stage `r2v` IDs, at the 768p class only (no two-stage ID). It is
+ *   subject to server authorization and needs `disableNSFWFilter: true`.
  * - The `i2v` model accepts `referenceImage`, `referenceImageEnd`, or both, and
  *   requires at least one of them. The `flf2v` model requires both.
  *
  * #### MiniMax H3 intermediate keyframes (every workflow except `t2v`)
  * - `keyframes` pins up to 8 still images at chosen frames between the first
- *   and last frame. 23 ids accept it (`isMinimaxH3KeyframeModel()`): `i2v` and
+ *   and last frame. 24 ids accept it (`isMinimaxH3KeyframeModel()`): `i2v` and
  *   `flf2v` on every tier (Standard, Balanced, LightX2V Turbo, FastH3 Turbo and
  *   FastH3 Two-Stage), the six FastH3 Sound to Video ids (`ia2v`, `flfa2v`,
- *   `a2v`, one- and two-stage) and the seven `r2v` ids. Text-to-video and
+ *   `a2v`, one- and two-stage) and the eight `r2v` ids. Text-to-video and
  *   every other model reject a non-empty list.
  * - Each entry is `{ image, frameIndex }`. `frameIndex` is the 0-based pixel
  *   frame at 24 fps (`Math.round(seconds * 24)`): an integer from 1 to
@@ -446,8 +451,9 @@ export interface MinimaxH3Keyframe {
  * - `minimax-h3-ref2va-fp8_r2v` (standard) and
  *   `minimax-h3-ref2va-fp8_r2v_turbo` (four-step Turbo), and
  *   `minimax-h3-ref2va-fp8_r2v_balanced` (eight-step Larry v4 Balanced) condition on labelled reference material
- *   rather than on frame anchors, as does FastH3 Reference
- *   (`minimax-h3-fastvideo-omni-int8_r2v_turbo`). The checkpoint accepts up to 9 reference
+ *   rather than on frame anchors, as do FastH3 Reference
+ *   (`minimax-h3-fastvideo-omni-int8_r2v_turbo`) and Prism
+ *   (`minimax-h3-prism-int8_r2v`). The checkpoint accepts up to 9 reference
  *   images, 3 reference videos (24fps, 2-15 seconds each), and 3 reference
  *   audio clips, with at most 12 reference files in total.
  * - All of those ceilings apply, and at least one visual reference (image or
@@ -510,14 +516,15 @@ export interface VideoProjectParams extends BaseProjectParams {
   /** Track one selected object across an uploaded video, preserving every source frame. */
   sam3VideoPrompt?: Sam3VideoPrompt;
   /**
-   * FlashVSR delivery resolution on the shorter edge. Source timing and audio are preserved.
-   * With a FlashVSR `referenceVideo`, `frames`, `fps`, `width` and `height` may be omitted: the
-   * server adopts the verified source's values. Values that are sent must match the source.
+   * FlashVSR or RelayVSR delivery resolution on the shorter edge; 2160 is RelayVSR only. Source
+   * timing and audio are preserved. With an upscale `referenceVideo`, `frames`, `fps`, `width` and
+   * `height` may be omitted: the server adopts the verified source's values. Values that are sent
+   * must match the source.
    */
-  upscaleResolution?: 1080 | 1440;
-  /** FlashVSR detail preference. Defaults to stable (More Stable). */
+  upscaleResolution?: 1080 | 1440 | 2160;
+  /** FlashVSR detail preference. Defaults to stable (More Stable). RelayVSR refuses it. */
   detailPreference?: 'stable' | 'sharper';
-  /** FlashVSR processing speed. Defaults to stable (More Stable). */
+  /** FlashVSR processing speed. Defaults to stable (More Stable). RelayVSR refuses it. */
   processingSpeed?: 'stable' | 'faster';
   /**
    * Number of frames to generate. Most requests pass `duration` instead and let
@@ -525,7 +532,7 @@ export interface VideoProjectParams extends BaseProjectParams {
    * it). Pass `frames` when positions inside the clip must be exact, as with
    * MiniMax H3 `keyframes`, whose `frameIndex` values must fit the count; H3
    * takes `124 + n*17` (124, 141, 158, ... 362). When both are passed,
-   * `duration` wins, except on FlashVSR upscales.
+   * `duration` wins, except on video upscales.
    */
   frames?: number;
   /**
@@ -697,7 +704,7 @@ export interface VideoProjectParams extends BaseProjectParams {
   /**
    * MiniMax H3 intermediate keyframes: up to 8 still images, each pinned at a
    * chosen frame between the first and last frame. Accepted by every MiniMax H3
-   * workflow except text-to-video, 23 ids (`isMinimaxH3KeyframeModel()`):
+   * workflow except text-to-video, 24 ids (`isMinimaxH3KeyframeModel()`):
    * image-to-video (`i2v`), first/last-frame (`flf2v`), Sound to Video (`ia2v`,
    * `flfa2v`, `a2v`) and Reference to Video (`r2v`). Any other model rejects a
    * non-empty list, and an empty list is the same as omitting the field.
@@ -1599,7 +1606,7 @@ export interface VideoEstimateRequest {
   includeMonthlyFairUse?: boolean;
   /** Include Priority mode's 1.5× usage in the daily-capacity estimate. */
   priorityMode?: boolean;
-  /** FlashVSR source geometry for an estimate. The server verifies the uploaded file at admission. */
+  /** Video upscale source geometry for an estimate. The server verifies the uploaded file at admission. */
   sourceWidth?: number;
   sourceHeight?: number;
   tokenType: TokenType;

package/src/Projects/utils/index.ts

@@ -71,7 +71,8 @@ const MINIMAX_H3_VIDEO_MODEL_IDS = new Set([
   'minimax-h3-ref2va-fp8_r2v_2stage',
   'minimax-h3-ref2va-fp8_r2v_balanced_2stage',
   'minimax-h3-fastvideo-omni-int8_r2v_turbo',
-  'minimax-h3-fastvideo-omni-int8_r2v_turbo_2stage'
+  'minimax-h3-fastvideo-omni-int8_r2v_turbo_2stage',
+  'minimax-h3-prism-int8_r2v'
 ]);
 
 export function getEnhacementStrength(strength: EnhancementStrength): number {
@@ -111,8 +112,18 @@ export function isVideoSegmentationModel(modelId: string): boolean {
 /** Standalone, promptless enhancement of a finished video. */
 export const FLASHVSR_VIDEO_UPSCALE_MODEL_ID = 'flashvsr_v1.1_tiny_long_bf16';
 
+/** RelayVSR: promptless upscaling at a fixed 2x or 4x, up to 2160p. */
+export const RELAYVSR_X2_VIDEO_UPSCALE_MODEL_ID = 'relayvsr_x2_lightx2v_int8_convrot';
+export const RELAYVSR_X4_VIDEO_UPSCALE_MODEL_ID = 'relayvsr_x4_int8_convrot';
+
+export function isRelayVsrModel(modelId: string): boolean {
+  return (
+    modelId === RELAYVSR_X2_VIDEO_UPSCALE_MODEL_ID || modelId === RELAYVSR_X4_VIDEO_UPSCALE_MODEL_ID
+  );
+}
+
 export function isVideoUpscaleModel(modelId: string): boolean {
-  return modelId === FLASHVSR_VIDEO_UPSCALE_MODEL_ID;
+  return modelId === FLASHVSR_VIDEO_UPSCALE_MODEL_ID || isRelayVsrModel(modelId);
 }
 
 /**
@@ -439,13 +450,20 @@ export function isWan3EnhancedModel(modelId: string): boolean {
  *   `minimax-h3-fastvideo-omni-int8_r2v_turbo_2stage` takes the same request
  *   and delivers the clip at exactly twice the canvas, like the Ref2VA
  *   two-stage ids.
+ * - Prism: `minimax-h3-prism-int8_r2v`, an optional reference-to-video model
+ *   for mature themes (a community merge of two Ref2VA finetunes) with its
+ *   recipe fixed at 8 Euler steps (see `isMinimaxH3PrismModel`). It takes the
+ *   same references, keyframes, catalog LoRAs and limits as the other one-stage
+ *   r2v ids at the 768p class only: it has no two-stage id. Server
+ *   authorization applies, and the Sensitive Content Filter must be off
+ *   (`disableNSFWFilter: true`).
  *
  * All H3 paths share fixed 24fps, guidance 1, and the `124 + n*17` frame grid.
  * Every path except the FastH3 audio guide generates 32kHz stereo audio
  * jointly; audio-guide output carries the uploaded audio instead. Standard H3 uses 20 steps;
  * Balanced uses qualified fixed 8-step acceleration: LightX2V for FL2VA and
  * Larry v4 for Ref2VA; each Turbo family uses its own 4-step distillation LoRA;
- * FastH3 Reference is distilled to a fixed 8 steps.
+ * FastH3 Reference is distilled to a fixed 8 steps, and so is Prism.
  */
 export function isMinimaxH3Model(modelId: string): boolean {
   return MINIMAX_H3_VIDEO_MODEL_IDS.has(modelId);
@@ -501,6 +519,15 @@ export function isMinimaxH3AudioGuideModel(modelId: string): boolean {
   return /^minimax-h3-fastvideo-int8_(?:ia2v|flfa2v|a2v)_turbo(?:_2stage)?$/.test(modelId);
 }
 
+/**
+ * Check if a model ID is MiniMax H3 Prism (`minimax-h3-prism-int8_r2v`): a
+ * one-stage reference-to-video id (`isMinimaxH3ReferenceModel`) whose steps are
+ * fixed at 8. It is not a Balanced id: it carries no Larry v4 adapter.
+ */
+export function isMinimaxH3PrismModel(modelId: string): boolean {
+  return modelId === 'minimax-h3-prism-int8_r2v';
+}
+
 /**
  * Check if a model ID is one of the 8-step MiniMax H3 Balanced workflows.
  * FL2VA covers t2v/i2v/flf2v; Ref2VA uses its matching Larry v4 adapter for
@@ -517,8 +544,9 @@ export function isMinimaxH3BalancedModel(modelId: string): boolean {
 /**
  * Check if a model ID is the MiniMax H3 Ref2VA multi-reference workflow
  * (`minimax-h3-ref2va-fp8_r2v`, `..._r2v_turbo`, `..._r2v_balanced`, the
- * two-stage `..._r2v_2stage` and `..._r2v_balanced_2stage`, or FastH3
- * Reference `minimax-h3-fastvideo-omni-int8_r2v_turbo` and its `_2stage` id).
+ * two-stage `..._r2v_2stage` and `..._r2v_balanced_2stage`, FastH3
+ * Reference `minimax-h3-fastvideo-omni-int8_r2v_turbo` and its `_2stage` id,
+ * or Prism `minimax-h3-prism-int8_r2v`).
  *
  * This is the only MiniMax H3 workflow that conditions on more than two input
  * files, and the only video workflow of any family that carries reference
@@ -541,13 +569,13 @@ export const MINIMAX_H3_MAX_KEYFRAMES = 8;
  * the worker pins (ComfyUI `MiniMaxH3AddGuide`) at chosen frames between the
  * first and last frame.
  *
- * Every MiniMax H3 workflow except text-to-video accepts them, 23 ids in all:
+ * Every MiniMax H3 workflow except text-to-video accepts them, 24 ids in all:
  * - image-to-video (`i2v`) and first/last-frame (`flf2v`) on every tier
  *   (Standard, Balanced, LightX2V Turbo, FastH3 Turbo and FastH3 Two-Stage);
  * - the FastH3 Sound to Video audio guide (`ia2v`, `flfa2v`, `a2v`, one-stage
  *   and two-stage; `isMinimaxH3AudioGuideModel`);
- * - Reference to Video (`r2v`: Standard, Turbo, Balanced, FastH3 Reference
- *   and their three two-stage ids; `isMinimaxH3ReferenceModel`).
+ * - Reference to Video (`r2v`: Standard, Turbo, Balanced, FastH3 Reference,
+ *   Prism and the three two-stage ids; `isMinimaxH3ReferenceModel`).
  */
 export function isMinimaxH3KeyframeModel(modelId: string): boolean {
   if (!isMinimaxH3Model(modelId)) return false;

package/src/index.ts

@@ -34,6 +34,8 @@ import {
   MINIMAX_H3_MAX_KEYFRAMES,
   PIXAL3D_IMAGE_TO_3D_MODEL_ID,
   PIXAL3D_MULTIVIEW_IMAGE_TO_3D_MODEL_ID,
+  RELAYVSR_X2_VIDEO_UPSCALE_MODEL_ID,
+  RELAYVSR_X4_VIDEO_UPSCALE_MODEL_ID,
   PIXAL3D_ORBIT_VIEW_SLOTS,
   SAM3_IMAGE_SEGMENT_MODEL_ID,
   SAM3_VIDEO_SEGMENT_MODEL_ID,
@@ -46,6 +48,7 @@ import {
   isModelArtifactModel,
   isPixal3dModel,
   isPixal3dMultiViewModel,
+  isRelayVsrModel,
   isSegmentationModel,
   isVideoModel,
   isVideoUpscaleModel,
@@ -562,6 +565,8 @@ export {
   PIXAL3D_ORBIT_VIEW_SLOTS,
   SAM3_IMAGE_SEGMENT_MODEL_ID,
   SAM3_VIDEO_SEGMENT_MODEL_ID,
+  RELAYVSR_X2_VIDEO_UPSCALE_MODEL_ID,
+  RELAYVSR_X4_VIDEO_UPSCALE_MODEL_ID,
   SEEDREAM_5_PRO_UNCENSORED_MODEL_ID,
   SEEDREAM_5_5_UNCENSORED_MODEL_ID,
   calculateVideoFrames,
@@ -573,6 +578,7 @@ export {
   isModelArtifactModel,
   isPixal3dModel,
   isPixal3dMultiViewModel,
+  isRelayVsrModel,
   isSegmentationModel,
   isSeedreamModel,
   isVideoModel,

package/src/lib/validation.ts

@@ -455,6 +455,7 @@ export function isComfyModel(modelId: string): boolean {
     'flux3_',
     'rtx_vsr_',
     'flashvsr_',
+    'relayvsr_',
     'wan_',
     'ace_step',
     'minimax_music3',

package/src/version.ts

@@ -1,3 +1,3 @@
 // AUTO-GENERATED by scripts/generate-version.mjs — do not edit.
 // Source of truth: package.json version.
-export const LIB_VERSION = '5.83.0';
+export const LIB_VERSION = '5.84.0';