AnimaV3PromptCheck:
AnimaV3PromptCheck checks a tag-based image prompt for rule violations and can automatically repair supported issues to suit a selected content level. It applies the plugin’s shared set of 17 fatal checks and 10 style checks, helping you identify conflicting instructions, formatting problems, and tags that exceed the intended content level before generating an image. Its repair capabilities cover removing over-level tags except registered calibration exceptions, resolving supported conflicts involving subject counts, viewpoints, eye coverings, clothing, equipment/depth-of-field combinations, and poses, and normalizing underscores and duplicate tags. It deliberately does not automatically remove unusually long tags that are absent from the vocabulary or reduce the total number of tags, because those decisions may require your artistic judgment. This is a rule-based prompt checker rather than an image-content guarantee. In the supplied V4 plugin, AnimaV3PromptCheck is retained as a compatibility alias for AnimaV4PromptCheck, so existing workflows using the V3 node type run the same implementation.
AnimaV3PromptCheck Input Parameters:
text
Required STRING containing the prompt you want to inspect, with an empty string as the default and a multiline editor for convenient entry. The node treats commas as tag separators, removes surrounding whitespace from each tag, and ignores empty entries between commas. For predictable results, supply comma-separated tags such as 1girl, solo, long hair, red dress; line breaks alone do not separate tags. When automatic repair is enabled, these tags may be modified by the supported repair rules before they are checked. When automatic repair is disabled, tag content is retained, but the text is still reconstructed with standardized comma-and-space separators. No minimum or maximum text length is declared in the interface, although the checker can report tag-length and tag-count issues.
level
Required COMBO selection defining the target content level used for checking and automatic repair, with sfw as the default. The node resolves the selected choice to an internal level key and its corresponding numeric level, then uses that target to identify violations and, when enabled, remove supported tags that exceed it. Select the level appropriate to your intended artwork rather than assuming the node will infer your desired restriction from the prompt. Registered calibration exceptions are excluded from the stated over-level tag removal behavior. The available choices are supplied by LEVEL_CHOICES, but their complete list and numeric bounds are not included in the provided source, so only the default sfw can be confirmed here.
auto_fix
Optional BOOLEAN controlling whether supported repairs are applied, with True as the default and True or False as the available values. Set it to True to repair supported content-level violations, mutually exclusive tags, underscore formatting, and duplicates before the node produces its check results. Set it to False when you want to inspect a prompt without applying those repairs, which is useful for preserving intentional combinations or understanding what needs attention before changing anything. Even with this option disabled, surrounding tag whitespace and comma spacing are normalized, and empty comma-separated entries are removed. Automatic repair does not shorten unusually long unknown tags or reduce the total tag count, so those issues may still require manual attention.
AnimaV3PromptCheck Output Parameters:
text
STRING containing the prompt after parsing and any enabled repairs, with tags joined using a comma followed by a space. When auto_fix is enabled, this is the repaired version that you can pass to a downstream text-processing or prompt-encoding node. When auto_fix is disabled, the tag content remains unchanged, but its separators and surrounding whitespace are normalized rather than preserved character for character. Empty or whitespace-only input produces an empty string. Review this output when exact wording matters, because supported repairs can remove tags or resolve conflicts in ways that change your intended composition.
report
STRING containing a formatted diagnostic report for the resulting tags, evaluated against the selected target level. The report formatter receives the check findings, the selected level’s display label, and any repair actions recorded during processing, allowing you to inspect both remaining issues and information about applied fixes. Because checking occurs after the repair step, enabling auto_fix means the findings describe the repaired prompt rather than an untouched copy of the original. Unusually long unknown tags and tag-count issues are not automatically fixed and can remain in the report. The supplied source does not expose the complete report template or the wording of every diagnostic.
level
STRING containing the resolved key for the target content level selected for this check. This identifies the restriction the node was asked to apply, not the content level it detected in the resulting prompt. For the default selection, the resolved target is sfw. Use this output when you want to retain or display the target alongside the checked prompt, and compare it with num when reviewing the distinction between the requested target and the checker’s assessed result.
num
INT containing the checker’s assessed content level for the resulting tags, taken from the report’s real_level value. It is calculated after any enabled repairs, so it describes the prompt returned by this node rather than necessarily describing the original prompt. Unlike level, which represents your selected target, num represents the rule-based assessment of the checked text. The supplied source does not define the numeric scale or its minimum and maximum values, so interpret the number using the plugin’s level definitions rather than assuming a particular scale.
AnimaV3PromptCheck Usage Tips:
- Begin with
levelset tosfwandauto_fixset toFalsewhen you want to understand the checker’s findings before permitting changes. Review the report, then enable repair if its supported changes suit your artwork. - Use comma-separated tags rather than a paragraph or newline-only list. For example,
1girl, solo, long hair, red dressgives the checker clearly separated entries. - Place the node after your final prompt-combination step and before prompt encoding so it checks the complete prompt you intend to use. Tags added downstream will not have been inspected by this node.
- Compare the original prompt with the returned
textwhenauto_fixis enabled, especially if your composition intentionally combines unusual clothing, viewpoints, or poses. - Review the report even after automatic repair. Long unknown tags and tag-count problems are intentionally left for you to resolve, and automatic repair should not be treated as proof that every check passed.
- If you need exact preservation of punctuation and spacing, keep a separate copy of the original text. Disabling
auto_fixpreserves tag content but does not bypass comma and whitespace normalization. - Use the
reportoutput with a compatible text-display node to make diagnostics visible while you work, and sendtextto the next prompt-processing stage. - Existing workflows can continue using the
AnimaV3PromptChecknode type with the supplied V4 plugin because it is registered as an alias of the V4 implementation.
AnimaV3PromptCheck Common Errors and Solutions:
Node type AnimaV3PromptCheck is missing
- Explanation: This is a workflow-loading symptom rather than a message generated by the checker. The supplied plugin registers the V3 name as a compatibility alias, so a missing node usually means the plugin is unavailable, failed to load, or is a different installation that does not register this alias.
- Solution: Confirm that Comfyui-Anima-Wardrobe-V4 is installed and loads successfully, restart ComfyUI, and inspect the startup log for plugin import errors. Use a version that includes the documented V3 compatibility mappings.
Tags above the selected content level remain in the prompt
- Explanation: This is a possible diagnostic outcome, not a confirmed literal error message. With
auto_fixdisabled, the checker does not remove over-level tags. With repair enabled, registered calibration exceptions are excluded from the stated over-level removal behavior, and the supplied source does not promise that every possible violation can be repaired. - Solution: Confirm that
auto_fixis enabled and thatlevelmatches your intended target. Read the report and manually revise remaining problematic tags rather than assuming automatic repair guarantees compliance.
Conflicting tags remain in the report
- Explanation: The node supports repairs for specific conflict categories, but disabling
auto_fixleaves conflicts in place, and combinations outside the supported rules may require manual editing. Exact diagnostic wording is not available in the supplied source. - Solution: Enable
auto_fixfor supported conflict resolution, then inspect the returned text and report. If a conflict remains, simplify the relevant subject count, viewpoint, eye covering, clothing, equipment/depth-of-field, or pose instructions to express one clear intention.
An unusually long unknown tag is reported but not repaired
- Explanation: This is intentional behavior. A long tag outside the plugin vocabulary could be a title or another meaningful phrase you wrote, so the node avoids automatically shortening or deleting it. The exact report message is not shown in the supplied source.
- Solution: Check whether the entry is intentional. Keep valid names or titles, shorten accidental prose, or add commas if several intended tags were mistakenly combined into one entry.
A tag-count issue remains after automatic repair
- Explanation: Automatic repair deliberately does not reduce tag count to satisfy a count check. Although other repairs may remove duplicates or conflicting tags, the node does not arbitrarily trim your prompt to meet a length target. Exact thresholds and message wording are not included in the supplied source.
- Solution: Review the report and manually adjust the number of tags. If the prompt comes from a generator with density or tag-limit controls, adjust those upstream settings before checking again.
Text formatting changes even when auto_fix is False
- Explanation: The node always splits text at commas, trims surrounding whitespace, discards empty entries, and rejoins tags with
", ". Turning off repair prevents rule-based changes to tag content, but it does not bypass this parsing and reconstruction. - Solution: Keep an original copy if exact formatting matters, and use the returned text when normalized comma-separated formatting is acceptable.
Newline-separated tags are treated as one entry
- Explanation: The parser separates entries only at commas. A list written with line breaks but no commas can therefore be checked as one combined tag rather than several independent tags.
- Solution: Insert commas between tags. You may retain line breaks for readability, but use commas wherever you intend a new tag to begin.
The text output is empty
- Explanation: Empty input, whitespace-only input, or input containing only commas produces no tags and therefore an empty text output. Enabled repairs can also remove supported tags, so the resulting prompt may contain fewer entries than you expected.
- Solution: Confirm that
textreceives a nonempty prompt. If the original contained tags, inspect the report and temporarily disableauto_fixto determine whether repairs removed them.
The level output does not match the number you expected
- Explanation:
levelandnumdescribe different things:levelis the selected target key, whilenumis the assessed numeric content level of the resulting tags. Their difference is not automatically an execution error, and the numeric mapping is not provided in the supplied source. - Solution: Use
levelto track your requested target and consult the report and plugin level definitions to interpretnum. If you intended to assess the original prompt, disableauto_fixbefore checking it.
