everypixel-cli 0.2.2__tar.gz → 0.4.0__tar.gz

{everypixel_cli-0.2.2 → everypixel_cli-0.4.0}/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## [0.4.0] - 2026-10-07
+
+- Add ElevenLabs speech synthesis and voice design.
+- Add saved voice creation, listing, deletion, and Qwen cloning by voice ID.
+
+## [0.3.0] - 2026-09-18
+
+- Add GPT Image 2.5 Sunburst and quality selection for GPT Image models.
+- Add GLM-5.3 chat with streaming and tool-call support.
+
 ## [0.2.2] - 2026-09-11
 
 - Increase the default task wait timeout to 30 minutes for long video generation.

{everypixel_cli-0.2.2 → everypixel_cli-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: everypixel-cli
-Version: 0.2.2
+Version: 0.4.0
 Summary: Official command line interface for Everypixel API
 Project-URL: Homepage, https://labs.everypixel.com
 Project-URL: Repository, https://github.com/everypixel/everypixel-cli
@@ -35,8 +35,8 @@ Description-Content-Type: text/markdown
 # Everypixel CLI
 
 Official command-line interface for the [Everypixel API](https://labs.everypixel.com/).
-Generate and edit images and videos, work with audio, run visual analysis, and
-automate Everypixel workflows from a terminal or an MCP client.
+Generate and edit images and videos, chat with GLM-5.3, work with audio, run
+visual analysis, and automate Everypixel workflows from a terminal or an MCP client.
 
 The CLI is designed for both people and agents: commands use readable terminal
 output by default and can return stable JSON for scripts and tool integrations.
@@ -181,6 +181,15 @@ to select their aspect ratio. Raster models also accept `--resolution`.
 
 `--lora-url` is supported only by `zimage`.
 
+GPT Image models are `gpt-image-2` and `gpt-image-2.5-sunburst`. Both generation
+and editing accept `--quality low|medium|high` (default: `medium`); Sunburst also
+supports `xhigh` and `max`. Use `--size square` with `--resolution 3k`.
+
+```bash
+everypixel image generate "a detailed botanical illustration" \
+  --model gpt-image-2.5-sunburst --quality high --download ./outputs
+```
+
 Grok image models are `grok-imagine`, `grok-imagine-2`, and
 `grok-imagine-2-low`. The `grok` alias remains supported. `grok_quality` and
 `wan22` have been removed from the API.
@@ -254,9 +263,12 @@ They support 4K output. The default `seedvr2` supports up to 1440p, with a
 | `everypixel lipsync video` | Synchronize a video with an audio track |
 | `everypixel lipsync image` | Create a lipsync video from a still image and audio |
 | `everypixel audio transcribe` | Transcribe speech from an audio file |
-| `everypixel audio tts-create` | Create speech with a built-in speaker |
-| `everypixel audio tts-clone` | Create speech from a cloned voice sample |
-| `everypixel audio tts-voice` | Create speech with a character voice |
+| `everypixel audio tts-create` | Create speech with a Qwen preset or saved ElevenLabs voice |
+| `everypixel audio tts-clone` | Create Qwen speech from an audio sample or saved voice |
+| `everypixel audio tts-design` | Design voice previews from a description |
+| `everypixel audio tts-voice` | Save audio samples or a design preview as a voice |
+| `everypixel audio tts-voices` | List presets and saved voices |
+| `everypixel audio tts-delete` | Delete an owned saved voice |
 
 Lipsync supports 720p; the API checks source media dimensions and duration.
 
@@ -281,6 +293,56 @@ everypixel audio tts-create \
   --download ./outputs
 ```
 
+`tts-create` lets the API choose the model when `--model` is omitted.
+`tts-clone` and `tts-design` default to `--model qwen3`.
+`--speaker`, `--style`, `--language`, and `--seed` are Qwen controls; `--character` applies only to Qwen voice design.
+Qwen text is limited to 200 characters. ElevenLabs synthesis accepts up to
+2048 characters and requires `--voice-id`. Supported synthesis models are
+`eleven_v4`, `eleven_v4_turbo`, `eleven_v3`, `eleven_v3_conversational`,
+`eleven_multilingual_v2`, and `eleven_flash_v2_5`. ElevenLabs design uses
+`eleven_multilingual_ttv_v2` or `eleven_ttv_v3`, with 100–1000 characters of
+text and a 20–1000 character `--prompt`.
+
+Save a returned preview with `tts-voice --name "Narrator" --preview-id UUID`, or save local
+samples with `tts-voice --name "My voice" --provider elevenlabs --audio ./sample.wav`.
+`--audio` is repeatable for ElevenLabs; Qwen (`--provider labs`, the default)
+requires one sample. Saving returns a voice card immediately. Use its ID with
+`tts-create --model eleven_v3 --voice-id UUID --text "Hello"` or
+`tts-clone --voice-id UUID --text "Hello"`. List voices with
+`tts-voices --provider elevenlabs --offset 0 --limit 50`, and delete an owned
+voice with `tts-delete --voice-id UUID`.
+
+### Chat
+
+```bash
+everypixel chat "Explain how a rainbow forms."
+everypixel chat "Review this idea." --system "Be concise." --stream
+everypixel chat "Solve this problem." --reasoning-effort high --show-reasoning
+everypixel chat --input-file ./conversation.json --output-json
+```
+
+Chat uses the same credentials as other commands and requires the client's
+`chat` scope. The default model is `glm-5.3`. Each invocation sends one request;
+use `--input-file` for conversation history, function tools, or other API options:
+
+```json
+{
+  "messages": [
+    {"role": "user", "content": "What causes rainbows?"},
+    {"role": "assistant", "content": "Sunlight refracts and reflects in water droplets."},
+    {"role": "user", "content": "Why do the colors separate?"}
+  ],
+  "max_completion_tokens": 4096
+}
+```
+
+A positional prompt appends a user message to the supplied history. CLI options
+override matching fields in the file. `--stream` displays text as it arrives;
+JSON output returns one assembled response, including reasoning, tool calls,
+and usage. Function calls are returned for the caller to handle. The MCP `chat`
+tool accepts the same structured request and returns a complete response.
+Use `--request-timeout` to change the chat HTTP timeout (default: 240 seconds).
+
 ### Visual Analysis
 
 Analysis commands accept HTTP(S) URLs and local files. Local files are uploaded

{everypixel_cli-0.2.2 → everypixel_cli-0.4.0}/README.md

@@ -1,8 +1,8 @@
 # Everypixel CLI
 
 Official command-line interface for the [Everypixel API](https://labs.everypixel.com/).
-Generate and edit images and videos, work with audio, run visual analysis, and
-automate Everypixel workflows from a terminal or an MCP client.
+Generate and edit images and videos, chat with GLM-5.3, work with audio, run
+visual analysis, and automate Everypixel workflows from a terminal or an MCP client.
 
 The CLI is designed for both people and agents: commands use readable terminal
 output by default and can return stable JSON for scripts and tool integrations.
@@ -147,6 +147,15 @@ to select their aspect ratio. Raster models also accept `--resolution`.
 
 `--lora-url` is supported only by `zimage`.
 
+GPT Image models are `gpt-image-2` and `gpt-image-2.5-sunburst`. Both generation
+and editing accept `--quality low|medium|high` (default: `medium`); Sunburst also
+supports `xhigh` and `max`. Use `--size square` with `--resolution 3k`.
+
+```bash
+everypixel image generate "a detailed botanical illustration" \
+  --model gpt-image-2.5-sunburst --quality high --download ./outputs
+```
+
 Grok image models are `grok-imagine`, `grok-imagine-2`, and
 `grok-imagine-2-low`. The `grok` alias remains supported. `grok_quality` and
 `wan22` have been removed from the API.
@@ -220,9 +229,12 @@ They support 4K output. The default `seedvr2` supports up to 1440p, with a
 | `everypixel lipsync video` | Synchronize a video with an audio track |
 | `everypixel lipsync image` | Create a lipsync video from a still image and audio |
 | `everypixel audio transcribe` | Transcribe speech from an audio file |
-| `everypixel audio tts-create` | Create speech with a built-in speaker |
-| `everypixel audio tts-clone` | Create speech from a cloned voice sample |
-| `everypixel audio tts-voice` | Create speech with a character voice |
+| `everypixel audio tts-create` | Create speech with a Qwen preset or saved ElevenLabs voice |
+| `everypixel audio tts-clone` | Create Qwen speech from an audio sample or saved voice |
+| `everypixel audio tts-design` | Design voice previews from a description |
+| `everypixel audio tts-voice` | Save audio samples or a design preview as a voice |
+| `everypixel audio tts-voices` | List presets and saved voices |
+| `everypixel audio tts-delete` | Delete an owned saved voice |
 
 Lipsync supports 720p; the API checks source media dimensions and duration.
 
@@ -247,6 +259,56 @@ everypixel audio tts-create \
   --download ./outputs
 ```
 
+`tts-create` lets the API choose the model when `--model` is omitted.
+`tts-clone` and `tts-design` default to `--model qwen3`.
+`--speaker`, `--style`, `--language`, and `--seed` are Qwen controls; `--character` applies only to Qwen voice design.
+Qwen text is limited to 200 characters. ElevenLabs synthesis accepts up to
+2048 characters and requires `--voice-id`. Supported synthesis models are
+`eleven_v4`, `eleven_v4_turbo`, `eleven_v3`, `eleven_v3_conversational`,
+`eleven_multilingual_v2`, and `eleven_flash_v2_5`. ElevenLabs design uses
+`eleven_multilingual_ttv_v2` or `eleven_ttv_v3`, with 100–1000 characters of
+text and a 20–1000 character `--prompt`.
+
+Save a returned preview with `tts-voice --name "Narrator" --preview-id UUID`, or save local
+samples with `tts-voice --name "My voice" --provider elevenlabs --audio ./sample.wav`.
+`--audio` is repeatable for ElevenLabs; Qwen (`--provider labs`, the default)
+requires one sample. Saving returns a voice card immediately. Use its ID with
+`tts-create --model eleven_v3 --voice-id UUID --text "Hello"` or
+`tts-clone --voice-id UUID --text "Hello"`. List voices with
+`tts-voices --provider elevenlabs --offset 0 --limit 50`, and delete an owned
+voice with `tts-delete --voice-id UUID`.
+
+### Chat
+
+```bash
+everypixel chat "Explain how a rainbow forms."
+everypixel chat "Review this idea." --system "Be concise." --stream
+everypixel chat "Solve this problem." --reasoning-effort high --show-reasoning
+everypixel chat --input-file ./conversation.json --output-json
+```
+
+Chat uses the same credentials as other commands and requires the client's
+`chat` scope. The default model is `glm-5.3`. Each invocation sends one request;
+use `--input-file` for conversation history, function tools, or other API options:
+
+```json
+{
+  "messages": [
+    {"role": "user", "content": "What causes rainbows?"},
+    {"role": "assistant", "content": "Sunlight refracts and reflects in water droplets."},
+    {"role": "user", "content": "Why do the colors separate?"}
+  ],
+  "max_completion_tokens": 4096
+}
+```
+
+A positional prompt appends a user message to the supplied history. CLI options
+override matching fields in the file. `--stream` displays text as it arrives;
+JSON output returns one assembled response, including reasoning, tool calls,
+and usage. Function calls are returned for the caller to handle. The MCP `chat`
+tool accepts the same structured request and returns a complete response.
+Use `--request-timeout` to change the chat HTTP timeout (default: 240 seconds).
+
 ### Visual Analysis
 
 Analysis commands accept HTTP(S) URLs and local files. Local files are uploaded

everypixel_cli-0.4.0/src/everypixel_cli/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.4.0"

everypixel_cli-0.4.0/src/everypixel_cli/application/chat.py

@@ -0,0 +1,192 @@
+"""Chat request construction and completion assembly without presentation code."""
+
+from __future__ import annotations
+
+import math
+from collections.abc import Callable, Iterator
+from typing import Any, Protocol
+
+from ..errors import APIResponseError, ValidationCLIError
+from ..schemas import ChatRequest
+from .models import ExecutionOptions, OperationResult
+
+
+class ChatClientProtocol(Protocol):
+    def request(
+        self,
+        method: str,
+        path: str,
+        *,
+        json: dict[str, Any] | None = None,
+        params: dict[str, Any] | None = None,
+        files: dict[str, Any] | None = None,
+        auth_required: bool = True,
+        timeout: float | None = None,
+    ) -> Any: ...
+
+    def stream_request(
+        self, method: str, path: str, *, json: dict[str, Any], timeout: float
+    ) -> Iterator[dict[str, Any]]: ...
+
+
+def build_chat_payload(
+    *,
+    payload: dict[str, Any],
+    prompt: str | None,
+    system: str | None,
+    options: dict[str, Any],
+) -> dict[str, Any]:
+    values = {
+        **payload,
+        **{key: value for key, value in options.items() if value is not None},
+    }
+    values.setdefault("model", "glm-5.3")
+    messages = values.get("messages", [])
+    if not isinstance(messages, list):
+        raise ValidationCLIError("messages must be an array")
+    messages = list(messages)
+    if system is not None:
+        messages.insert(0, {"role": "system", "content": system})
+    if prompt is not None:
+        messages.append({"role": "user", "content": prompt})
+    values["messages"] = messages
+    if values.get("stream") and "stream_options" not in values:
+        values["stream_options"] = {"include_usage": True}
+    return ChatRequest.model_validate(values).model_dump(mode="json", exclude_none=True)
+
+
+class ChatCompletion:
+    """Combine text, reasoning, and tool-call fragments into a completion."""
+
+    def __init__(self) -> None:
+        self.metadata: dict[str, Any] = {}
+        self.choices: dict[int, dict[str, Any]] = {}
+        self.tools: dict[int, dict[int, dict[str, Any]]] = {}
+
+    def add(self, event: dict[str, Any]) -> None:
+        choices = event.get("choices")
+        if not isinstance(choices, list):
+            raise APIResponseError("API returned invalid chat choices")
+        self.metadata.update(
+            {key: value for key, value in event.items() if key != "choices"}
+        )
+        for item in choices:
+            if not isinstance(item, dict):
+                raise APIResponseError("API returned an invalid chat choice")
+            index = item.get("index")
+            delta = item.get("delta")
+            if type(index) is not int or index < 0 or not isinstance(delta, dict):
+                raise APIResponseError("API returned an invalid chat delta")
+            choice = self.choices.setdefault(
+                index,
+                {
+                    "index": index,
+                    "message": {"role": "assistant", "content": None},
+                    "finish_reason": None,
+                },
+            )
+            choice.update(
+                {
+                    key: value
+                    for key, value in item.items()
+                    if key not in {"delta", "finish_reason"}
+                }
+            )
+            if item.get("finish_reason") is not None:
+                choice["finish_reason"] = item["finish_reason"]
+            message = choice["message"]
+            for key in ("content", "reasoning_content", "refusal"):
+                append_fragment(message, delta, key)
+            if "role" in delta:
+                message["role"] = delta["role"]
+            tool_calls = delta.get("tool_calls")
+            if tool_calls is not None:
+                self._add_tools(index, tool_calls)
+
+    def _add_tools(self, choice_index: int, calls: Any) -> None:
+        if not isinstance(calls, list):
+            raise APIResponseError("API returned invalid tool calls")
+        tools = self.tools.setdefault(choice_index, {})
+        for call in calls:
+            if not isinstance(call, dict):
+                raise APIResponseError("API returned an invalid tool call")
+            index = call.get("index")
+            if type(index) is not int or index < 0:
+                raise APIResponseError("API returned an invalid tool call index")
+            tool = tools.setdefault(index, {"function": {"name": "", "arguments": ""}})
+            append_fragment(tool, call, "id")
+            if "type" in call:
+                tool["type"] = call["type"]
+            function = call.get("function")
+            if function is not None:
+                if not isinstance(function, dict):
+                    raise APIResponseError("API returned an invalid function call")
+                for key in ("name", "arguments"):
+                    append_fragment(tool["function"], function, key)
+
+    def result(self) -> dict[str, Any]:
+        if not self.choices or any(
+            choice["finish_reason"] is None for choice in self.choices.values()
+        ):
+            raise APIResponseError("API returned an unfinished chat completion")
+        for index, tools in self.tools.items():
+            self.choices[index]["message"]["tool_calls"] = [
+                tools[key] for key in sorted(tools)
+            ]
+        return {
+            **self.metadata,
+            "object": "chat.completion",
+            "choices": [self.choices[index] for index in sorted(self.choices)],
+        }
+
+
+def append_fragment(target: dict[str, Any], delta: dict[str, Any], key: str) -> None:
+    value = delta.get(key)
+    if value is None:
+        return
+    if not isinstance(value, str):
+        raise APIResponseError("API returned an invalid chat text fragment")
+    target[key] = (target.get(key) or "") + value
+
+
+class ChatService:
+    def __init__(self, client: ChatClientProtocol) -> None:
+        self.client = client
+
+    def execute(
+        self,
+        payload: dict[str, Any],
+        *,
+        request_timeout: float = 240.0,
+        execution: ExecutionOptions | None = None,
+        on_chunk: Callable[[dict[str, Any]], None] | None = None,
+    ) -> OperationResult:
+        if not math.isfinite(request_timeout) or request_timeout <= 0:
+            raise ValidationCLIError("request timeout must be a finite positive number")
+        execution = execution or ExecutionOptions()
+        execution.check_cancelled()
+        if not payload.get("stream"):
+            value = self.client.request(
+                "POST", "/v1/chat/completions", json=payload, timeout=request_timeout
+            )
+            execution.check_cancelled()
+            return OperationResult(value=value)
+        completion = ChatCompletion()
+        events = self.client.stream_request(
+            "POST", "/v1/chat/completions", json=payload, timeout=request_timeout
+        )
+        # Closing the generator also closes its HTTP response on cancellation
+        # or a rendering failure, without starting another generation.
+        try:
+            for event in events:
+                execution.check_cancelled()
+                completion.add(event)
+                if on_chunk is not None:
+                    on_chunk(event)
+                execution.check_cancelled()
+        finally:
+            close = getattr(events, "close", None)
+            if close is not None:
+                close()
+        execution.check_cancelled()
+        return OperationResult(value=completion.result())