ATH-MaaS/Pixelle-Video · error · KeyError

Invalid response format: missing 'image_prompts'

Error message

Invalid response format: missing 'image_prompts'

What it means

generate_image_prompts batches narrations and asks an LLM to return JSON containing an 'image_prompts' key. When the parsed LLM response lacks that key, a KeyError('Invalid response format: missing 'image_prompts'') is raised so the retry loop can regenerate the batch. It indicates the model deviated from the required JSON schema or the JSON parse produced an object without the field.

Solutions

1. Retry the call — generate_image_prompts already loops max_retries times; transient model drift usually resolves on retry.
2. Inspect the raw LLM response (logged at debug level) to see the actual JSON shape and adjust the prompt to explicitly demand {"image_prompts": [...]}.
3. Reduce batch size so the model reliably emits one prompt per narration.
4. Switch to a model/provider that follows structured output or use JSON-mode/forced-schema output.
5. Patch _parse_json/callers to accept alternative key names or a top-level array.

Example fix

// before
result = _parse_json(response)
if "image_prompts" not in result:
    raise KeyError("Invalid response format: missing 'image_prompts'")
// after
result = _parse_json(response)
if "image_prompts" not in result:
    if isinstance(result, list):  # model returned a bare array
        result = {"image_prompts": result}
    else:
        raise KeyError("Invalid response format: missing 'image_prompts'")

def looks_like_image_prompt_response(payload) -> bool:
    return isinstance(payload, dict) and isinstance(payload.get("image_prompts"), list) and len(payload["image_prompts"]) > 0
# call after _parse_json, before consuming result

def has_image_prompts(obj: object) -> bool:
    return isinstance(obj, dict) and isinstance(obj.get("image_prompts"), list) and all(isinstance(p, str) for p in obj["image_prompts"])

try:
    prompts = generator.generate_image_prompt(narrations)
except KeyError as e:
    if "image_prompts" in str(e):
        logger.warning("LLM ignored schema, retrying with smaller batch")
        prompts = generator.generate_image_prompt(narrations[: len(narrations) // 2])
    else:
        raise

When it happens

Trigger: The LLM returns valid JSON but without an 'image_prompts' array — e.g. the model wrapped output in prose, returned prompts under a different key, returned a bare array, or a retry consumed a malformed response after _parse_json salvaged a partial object.

Common situations: Weaker/cheaper LLM models ignoring the JSON schema instruction; prompts asking for many narrations in one batch causing truncated or restructured output; model updates changing output shape; custom system prompts overriding the format instructions.

Related errors

• Invalid response format: missing 'narrations' key
• Invalid response format: missing 'video_prompts'
• Failed to parse LLM response as
• Batch prompt count mismatch (attempt / ): Expected: prompts…
• DashScope failed after attempts due to network error

AI-assisted analysis of ATH-MaaS/Pixelle-Video@848b054e4f (2026-08-30). Data as JSON: /api/errors/649f72ed5b223a29. Report an issue: GitHub.

Appendix: source

Thrown at pixelle_video/utils/content_generators.py:329

                prompt = build_image_prompt_prompt(
                    narrations=batch_narrations,
                    min_words=min_words,
                    max_words=max_words
                )
                
                response = await llm_service(
                    prompt=prompt,
                    temperature=0.7,
                    max_tokens=8192
                )
                
                logger.debug(f"Batch {batch_idx} attempt {attempt}: LLM response length: {len(response)} chars")
                
                # Parse JSON
                result = _parse_json(response)
                
                if "image_prompts" not in result:
                    raise KeyError("Invalid response format: missing 'image_prompts'")
                
                batch_prompts = result["image_prompts"]
                
                # Validate count
                if len(batch_prompts) != len(batch_narrations):
                    error_msg = (
                        f"Batch {batch_idx} prompt count mismatch (attempt {attempt}/{max_retries}):\n"
                        f"  Expected: {len(batch_narrations)} prompts\n"
                        f"  Got: {len(batch_prompts)} prompts"
                    )
                    logger.warning(error_msg)
                    
                    if attempt < max_retries:
                        logger.info(f"Retrying batch {batch_idx}...")
                        continue
                    else:
                        raise ValueError(error_msg)
                

View on GitHub (pinned to 848b054e4f)