AnimaV3PromptGrader:
AnimaV3PromptGrader reviews a text prompt and assigns an adult-content or NSFW classification using the Anima Wardrobe rule tables. It also identifies recognized adult-content categories, helping you understand why a prompt received its classification before you use it for image generation. This is a rule-based review of prompt tags, not an inspection of generated images or a guarantee about what an image model will produce. It shares its classification tables with the Anima random-prompt generator, keeping grading consistent between generation and later review. The node does not rewrite prompts, remove sensitive terms, or repair conflicting tags. In the supplied V4 package, AnimaV3PromptGrader is retained as a backward-compatible alias of AnimaV4PromptGrader, allowing older workflows to use the current implementation without replacing the saved node type.
AnimaV3PromptGrader Input Parameters:
text
A required STRING containing the prompt you want to classify. The default is an empty string, and the interface supports multiline text; no minimum or maximum length is specified. The node splits this text at commas, removes spaces from the beginning and end of each resulting tag, and ignores empty entries before grading. For example, 1girl, solo, long hair becomes three tags for analysis. Line breaks alone do not separate tags, so comma-separated prompts are the clearest format for this node. The recognized tags determine the classification, category matches, and supporting information in the report. These preparation steps apply only to analysis: the returned prompt text retains its original formatting.
detailed
An optional BOOLEAN that controls whether the report includes an additional list of matched classification tags. Its default is True, and its available options are True and False; numeric minimum and maximum values do not apply. When enabled, the node adds a “命中标签:” line if the classifier returns any matching tags, showing at most 20 of them. When disabled, that extra line is omitted, but the report still includes the classification, baseline explanation, tag count, and any recognized adult-content categories. This setting changes the amount of supporting detail shown, not the classification result or the prompt text.
AnimaV3PromptGrader Output Parameters:
text
A STRING containing the original prompt text, unchanged for normal text input. Although the node trims and separates tags internally for analysis, it does not apply those changes to this output. This lets you pass your prompt onward while reviewing its classification separately. Importantly, this output is not a cleaned, filtered, or safer version of the prompt: any sensitive terms, repeated tags, or formatting issues remain present.
num
An INT containing the numeric classification assigned by the shared Anima NSFW rules. This gives you a machine-readable value for comparing results or connecting to compatible downstream logic. The report displays this number alongside the classifier’s maximum scale value, providing context for interpreting it. The supplied node code does not define the complete numeric range or the meanings of individual thresholds, so you should consult the report and the installed classification rules rather than assume a particular rating scale.
level
A STRING containing the classifier’s level identifier, taken directly from its classification result. It represents the same classification as num in a text-based form and can be useful when displaying results or connecting to compatible text-processing nodes. The report shows this identifier in parentheses after a separate readable classification label. The supplied source does not enumerate the possible identifiers, so their exact values should not be assumed.
report
A STRING containing a multiline explanation of the classification, with headings written in Chinese. It begins with “档位” for the numeric classification, maximum scale value, readable label, and level identifier; “基线” for the classifier’s baseline explanation; and “标签” for the number of nonempty comma-separated entries analyzed. It then lists recognized adult-content categories under “成人向内容”, or displays “成人向内容:无” when none are returned. Each listed category includes its item count and up to six matching entries, followed by an ellipsis if more exist. If detailed is enabled and matching tags are available, a final “命中标签:” line shows up to 20 matches. This report helps you interpret the result, but its displayed examples may not include every match.
AnimaV3PromptGrader Usage Tips:
- Separate prompt tags with commas. For example, use
1girl, solo, long hairrather than placing each term on a separate line without commas; the node uses commas, not line breaks, to divide the text for analysis. - Keep
detailedset toTruewhen investigating an unexpected classification. The additional matched-tag line can help you identify which terms contributed to the result. - Set
detailedtoFalsewhen you want a shorter report for routine checks. This removes only the extra matched-tag line; it does not change the grade or hide the category summaries. - Review
reportalongsidenumandlevel. A numeric grade alone does not explain the classifier’s reasoning, and the complete scale definitions are not included in the supplied node code. - Grade the final assembled prompt if you want to assess all the text you intend to send to image generation. Any terms added after this review will not be included in its classification.
- Use AnimaV4PromptCheck when you need diagnosis and supported automatic repairs rather than classification alone. AnimaV3PromptGrader leaves the original text unchanged, even when it recognizes adult-content terms.
- Treat the result as a prompt-review aid, not a guarantee about generated images. The node analyzes text using registered rules and does not inspect image content.
- Keep the V3 node in an existing workflow when using the supplied V4 package if it loads correctly. The package explicitly registers the V3 name as an alias of the V4 implementation.
AnimaV3PromptGrader Common Errors and Solutions:
Missing node: AnimaV3PromptGrader
- Explanation: This is a workflow-loading issue rather than an error raised by the grading method. The supplied package registers the V3 node name for compatibility, so a missing-node warning can indicate that the package is absent, failed to load, or differs from the supplied version.
- Solution: Confirm that Comfyui-Anima-Wardrobe-V4 is installed and enabled, restart ComfyUI, and check its startup log for loading errors. Use a package version that includes the V3 alias registration before changing older workflows.
The report shows “标签 : 0 个”
- Explanation: No nonempty tags were found. An empty prompt, whitespace-only text, or text consisting only of commas produces zero analyzed tags. The node does not explicitly reject this input before calling the classifier.
- Solution: Enter or connect the prompt you intended to review, then run the node again. Do not treat an empty-input classification as a meaningful assessment of your intended prompt.
The tag count is lower than expected
- Explanation: The node splits text only at commas and ignores empty entries. Multiple terms separated only by line breaks remain in the same entry, while repeated commas do not create additional tags.
- Solution: Add commas between the tags you want analyzed separately. For example, change a three-line list into
portrait, solo, long hairand run the review again.
The report shows “成人向内容:无” despite concerning wording
- Explanation: This means the classifier returned no adult-content category entries; it is not an execution error or proof that the prompt is suitable for every audience. This node uses registered classification rules rather than an unrestricted language-based review.
- Solution: Check your tag formatting and terminology, review the baseline explanation, and inspect the prompt yourself. Use the result as supporting information rather than a complete content-safety assessment.
The “命中标签:” line is missing
- Explanation: The node adds this line only when
detailedisTrueand the classifier returns a nonempty list of matching tags. Category information can still appear independently of this additional line. - Solution: Enable
detailedand run the node again. If the line remains absent, consult the baseline and category sections; the node may simply have no matching tags to list there.
Some matching tags are missing from the report
- Explanation: The report intentionally limits its displayed examples. Each category shows up to six entries, and the optional matched-tag line shows up to 20. An ellipsis after a category’s entries indicates that additional entries exist.
- Solution: Use the category counts to recognize when the displayed list is incomplete. If you need to investigate individual terms, review smaller portions of the prompt separately, then grade the full prompt again for its overall result.
The prompt is unchanged after grading
- Explanation: This is expected behavior. The node classifies the prompt but does not remove adult-content terms, normalize the returned text, or resolve conflicting tags.
- Solution: Edit the prompt yourself or use AnimaV4PromptCheck for its supported diagnosis and repair features. Run the grader again after making changes.
The classification does not match the generated image
- Explanation: The node reviews prompt text only. It does not inspect the generated image, and its rule-based classification cannot guarantee how an image model interprets the prompt.
- Solution: Review the generated image separately and confirm that you graded the complete final prompt. Adjust your prompt as needed rather than relying on the text grade alone.
