Anima 提示词检查与修复(V4):
AnimaV4PromptCheck checks comma-separated image prompts for rule violations and can automatically repair supported problems according to a chosen content level. It applies the Anima rule system’s 17 fatal checks and 10 style checks, helping you identify inconsistent tags, formatting issues, and content that exceeds your intended level before generating an image. Its automatic repairs cover three main areas: removing over-level tags except registered calibration exceptions, resolving supported conflicts involving subject counts, viewpoints, eye coverings, clothing, equipment/depth of field, and poses, and normalizing underscores and removing duplicates. The node uses the same repair routine as the prompt-generation process, keeping these corrections consistent across the plugin. It deliberately leaves certain issues for your judgment: unusually long tags that are absent from the vocabulary may be intentional titles, and excessive tag counts should be addressed by adjusting prompt density or limits rather than indiscriminately deleting content. This is a text-checking tool, not an image inspection tool or a guarantee of what a model will generate, and it only affects a workflow when you connect and use it.
Anima 提示词检查与修复(V4) Input Parameters:
text
Required STRING containing the prompt you want to inspect, with a default of an empty string and a multiline text editor. The node treats commas as tag separators, trims surrounding whitespace from each tag, and ignores empty entries; for example, portrait, blue eyes,, soft lighting becomes three tags. Line breaks alone are not used as tag separators, so enter comma-separated tags even when arranging the prompt over several lines. When automatic repair is enabled, supported tags may be removed, normalized, or deduplicated before checking; when it is disabled, those repairs are skipped, although the basic splitting and spacing cleanup still occurs. No minimum or maximum character count is specified in the interface, and an empty prompt is accepted by the method rather than explicitly rejected.
level
Required COMBO selection that sets the target content level used for checking and, when enabled, automatic repair. The default is sfw. The available selections come from LEVEL_CHOICES, but their complete list is not included in the supplied source, so additional option names and their numeric limits cannot be reliably listed here. Choose the level that matches your intended content rather than assuming the prompt already meets it: the node resolves your selection to its internal level key and numeric threshold, then uses that threshold to identify violations and remove supported over-level tags. Registered calibration exceptions are excluded from the documented over-level removal behavior. This setting expresses your target; it does not by itself prove that the resulting prompt falls within that target.
auto_fix
Optional BOOLEAN that defaults to True and supports True or False; numeric minimum and maximum values do not apply. With True, a nonempty tag list is passed through the shared repair routine before the node performs its checks, allowing supported content-level, conflict, and formatting problems to be corrected automatically. With False, the repair routine is skipped so you can inspect problems without removing or replacing tags through that routine. However, this is not an exact character-for-character passthrough: the node still trims tag edges, drops empty comma-separated entries, and rejoins the tags with a comma followed by a space. Regardless of this setting, excessively long unknown tags and tag-count problems are not automatically fixed.
Anima 提示词检查与修复(V4) Output Parameters:
text
STRING containing the tag list after processing, rejoined using , between tags. With automatic repair enabled, this is the repaired prompt; with automatic repair disabled, it retains the supplied tags apart from basic separator and surrounding-whitespace cleanup. This is the text you can connect to your prompt encoder or another text-processing node when you want subsequent steps to use the checked version. It may still contain reported problems that the repair routine intentionally does not address, so inspect the report rather than treating this output as an unconditional approval.
report
STRING containing a formatted diagnostic report produced from the processed tags and the selected target level, with repair information supplied when automatic changes were made. Because checking happens after the repair step, the diagnostic results describe the resulting prompt rather than a separate, untouched copy of the original. Use this output to understand remaining issues and review the available record of automatic fixes. The report is especially important for problems that are deliberately left unresolved, including overly long tags outside the vocabulary and tag-count issues. The exact report wording and layout are not included in the supplied code.
level
STRING containing the resolved internal key for the selected target level. This identifies the level the node was asked to check against, not an independent classification of the resulting text. For example, the default selection resolves to the target key used for sfw. This output is useful when you want to display or pass along the target setting together with the checked prompt; compare it with the report and the actual-level number instead of interpreting it as confirmation that all checks passed.
num
INT containing the report’s real_level value, converted to an integer. It represents the actual level determined by the rule checker for the processed tags, after repairs if they were enabled. Unlike level, which records your target, this output reflects the checker’s assessment of the resulting prompt. The supplied code does not include the complete numeric-to-level mapping or numeric range, so interpret the number alongside the readable report rather than assigning unsupported meanings to specific values. It is a rule-based prompt classification, not an assessment of a generated image.
Anima 提示词检查与修复(V4) Usage Tips:
- Start with
auto_fix=Falsewhen reviewing a carefully written prompt. This lets you see the diagnosis before allowing supported tag removals or corrections; keep a copy of the original if exact formatting matters. - Use
auto_fix=Truewhen preparing prompts for generation and choose the targetlevelfirst. Review bothtextandreportafterward to confirm that the repairs preserved your intended composition. - Write tags with commas between them. For example, use
portrait, blue eyes, soft lightingrather than relying on line breaks to separate three tags. - Connect the
textoutput to the prompt encoder if you want the checked or repaired prompt to affect generation. Viewing the report alone does not replace text already supplied through another connection. - Read the report after automatic repair instead of assuming all problems have disappeared. Long unknown tags and tag-count issues require your own review even when
auto_fix=True. - If a tag-count issue appears, reduce unnecessary descriptors manually or adjust the density or tag limit in the upstream generator, where those controls are available. This node does not expose a tag-count limit.
- Use AnimaV4PromptGrader when you only need content-level and category assessment; use AnimaV4PromptCheck when you also need rule diagnostics and supported repairs.
Anima 提示词检查与修复(V4) Common Errors and Solutions:
Tags disappear after checking
- Explanation: This is a possible processing result, not a quoted exception message. With
auto_fix=True, the repair routine can remove over-level tags, eliminate duplicates, and resolve supported conflicts. - Solution: Review the repair information in
reportand compare the result with your original prompt. Setauto_fix=Falseto inspect the prompt without automatic repairs, or revise conflicting tags yourself.
Spacing changes even with auto_fix=False
- Explanation: Disabling automatic repair does not disable basic text parsing. The method always splits on commas, trims tag edges, removes empty entries, and rejoins tags using a comma and a space.
- Solution: Treat this mode as “no rule-based repairs,” not exact text preservation. Keep the original text separately if you need its formatting unchanged.
Long or unknown tags remain after automatic repair
- Explanation: The node explicitly does not automatically fix tags that are both unusually long and absent from its vocabulary, because they might be intentional titles or other custom text. This is a documented limitation, not necessarily a processing failure.
- Solution: Inspect the reported tag and keep it if intentional. Otherwise, shorten it, correct it, or separate accidentally combined tags with commas.
A tag-count issue remains after automatic repair
- Explanation: Tag-count problems are reported but are not automatically repaired. The node avoids arbitrarily cutting your prompt down to a smaller number of tags.
- Solution: Remove less important tags manually or adjust density and tag-cap controls in the upstream generator, if available. There is no tag-count input on this node.
Multiline text is not treated as separate tags
- Explanation: The parser splits only on commas. A line break without a comma does not create another tag, which can leave several descriptions grouped into one entry.
- Solution: Add commas between intended tags. You can retain line breaks for readability as long as the comma separators are present.
The selected level does not match the assessed level
- Explanation: The
leveloutput identifies the selected target, whilenumrepresents the checker’s actual-level assessment of the processed prompt. These serve different purposes, and selecting a target does not guarantee compliance, especially when automatic repair is disabled. - Solution: Read the report to understand the remaining issues. Enable automatic repair for supported corrections, then manually review anything that remains unresolved.
The image still uses the original prompt
- Explanation: This node does not automatically replace prompts elsewhere in your workflow. Running it for inspection will not make an encoder use its processed text unless the connection is present.
- Solution: Connect the node’s
textoutput to the intended prompt input and check that no separate text path is supplying the original prompt instead.
The checked text output is empty
- Explanation: Empty input, whitespace-only input, or comma-only input produces no tags during parsing. Automatic repair may also remove tags; the supplied implementation does not explicitly reject an empty result.
- Solution: Check the original prompt and the report. Provide meaningful comma-separated tags, or temporarily disable automatic repair to determine whether repairs removed the content.
Exact exception or report message is not documented
- Explanation: The supplied node method does not define explicit exception messages, and the implementations of its report formatter and shared rules are not provided. Specific runtime error strings or exact diagnostic wording therefore cannot be verified from this context.
- Solution: Use the actual report or ComfyUI error text when troubleshooting. Check the prompt format and selected options first, and include the exact message, plugin version, and a minimal prompt example when requesting support.
