calesthio/OpenMontage · error · ValueError

Gemini Omni video_edit accepts at most 10 reference images

Error message

Gemini Omni video_edit accepts at most 10 reference images

What it means

Raised in _build_payload for gemini_video_edit when the number of reference images exceeds the cap read from spec['media_limits'].get('images', 10). The hardcoded message says 10, matching the default; if a spec declares a lower cap the check uses that number but the message text still says 10.

Source

Thrown at tools/video/atlas_video.py:292

            if not refers or not any(item.get("type") in {"image", "video"} for item in refers):
                raise ValueError("MiniMax H3 reference_to_video requires at least one image or video in refers")
            payload["refers"] = refers
        elif style == "gemini_video_clips":
            clips = list(inputs.get("video_clips") or [])
            if not clips and video:
                clips = [{"url": video, "start": 0, "ends": min(int(inputs.get("duration", 10)), 10)}]
            if len(clips) != 1:
                raise ValueError("Gemini Omni developer reference_to_video requires exactly one video_clip")
            payload["video_clips"] = clips
            if images:
                payload["images"] = images
        elif style == "gemini_video_edit":
            if not video:
                raise ValueError("video_edit requires video_url, video_path, or reference_video_path")
            payload["video"] = video
            if images:
                if len(images) > spec["media_limits"].get("images", 10):
                    raise ValueError("Gemini Omni video_edit accepts at most 10 reference images")
                payload["images"] = images

        extra = inputs.get("extra_params")
        if isinstance(extra, dict):
            payload.update(extra)
        return payload

    @staticmethod
    def _upload_value(value: str | None, api_key: str) -> str | None:
        if not value or _is_remote(value):
            return value
        return atlas_client.upload_media(value, api_key)

    def _resolve_media(self, inputs: dict[str, Any], api_key: str) -> dict[str, Any]:
        resolved = dict(inputs)
        resolved["reference_images"] = [
            *(resolved.get("reference_images") or []),
            *(resolved.get("reference_image_urls") or []),

View on GitHub (pinned to 95e1c3d0ab)

Solutions

1. Trim reference_images to at most 10 (or the spec's media_limits['images'] if lower — verify via get_info())
2. Prefer fewer, high-impact reference images for edit guidance
3. If you need more references, use a reference_to_video model with higher media_limits

Example fix

# before
inputs = {'model':'google/gemini-omni-flash','operation':'video_edit','video_url':v,'reference_images':frames}  # len(frames)=25

# after
inputs = {'model':'google/gemini-omni-flash','operation':'video_edit','video_url':v,'reference_images':frames[:10]}

cap = atlas_video.get_info()['model_catalog'][model]['media_limits'].get('images', 10)
imgs = inputs.get('reference_images') or []
if len(imgs) > cap:
    inputs['reference_images'] = imgs[:cap]  # keep the strongest refs first

def within_edit_image_cap(inputs: dict, cap: int = 10) -> bool:
    return len(inputs.get('reference_images') or []) <= cap

try:
    result = atlas_video.run(inputs=inputs)
except ValueError as e:
    if 'video_edit accepts at most' in str(e):
        inputs['reference_images'] = inputs['reference_images'][:10]
        result = atlas_video.run(inputs=inputs)
    else:
        raise

When it happens

Trigger: Passing 11+ reference_images alongside the edit video on the Gemini Omni route; a spec defining media_limits['images'] below 10 while the caller sends 10 images (check fails, message misleadingly claims the cap is 10).

Common situations: Batch pipelines attaching every storyboard frame as an edit reference; porting a seedance multi-reference call (which allows far more) to the Gemini edit route.

Related errors

• video_edit requires video_url, video_path, or reference_vide
• This Gemini route requires at least one reference image
• {model} accepts at most {limits['images']} images, {limits['
• Gemini Omni developer reference_to_video requires exactly on
• Input video not found: {path}

AI-assisted analysis of calesthio/OpenMontage@95e1c3d0ab (2026-08-15). Data as JSON: /api/errors/49e8de1f99927a10. Report an issue: GitHub.