Eric HissandClaude Fable 5.1
4459af9ea6
Every node id, input name and existing widget position is unchanged, so saved
workflows load; new inputs are appended only.
Wildcards
- core/wildcards: Dynamic Prompts / Impact Pack grammar, tag queries, variables,
conditionals, concept / verbatim / loose bindings, reproducible seeds.
Wired into every enhancer node; reads ComfyUI/models/wildcards.
- New nodes: Wildcard Processor, Semantic Variation (direction axes),
Concept Survival Check. Starter lists in wildcards/.
- Random choices use a private generator; global random state is untouched.
Platform profiles
- core/profiles + profiles/*.yaml: one file per model (format, length,
negative prompt or none, quality tokens, text and dialogue syntax, vendor
rewriter prompt fetched once and cached). profiles_user/ overrides by id.
- New node: Platform Prompt Rewriter (prose, tags, edit instructions, JSON,
MiniMax H3 fields, Hailuo brackets), with validators for each format.
- Text-to-Image and Image-to-Image read profiles through platforms.py;
corrected facts for FLUX, Kontext, Chroma, Wan, AuraFlow, Kolors, NoobAI,
Qwen-Image, HunyuanImage. Negative prompt is empty for models without one.
- Hosted / API models: Seedream 5, Nano Banana, GPT Image, Midjourney V8,
FLUX.2 pro, Ideogram 3, Recraft V4, Seedance 2.0 / 2.5, Kling 3.0, Veo 3.1,
Gemini Omni Flash, Runway, Wan 3.0, Luma, hosted H3, Grok, Vidu, PixVerse.
Reports say how to switch the vendor's own prompt rewriting off.
- Illustrious: spaced tags, quality tags last, 25-80 words (OnomaAI paper).
Video nodes
- target_model and duration_seconds appended to the three video expanders.
Default target is MiniMax H3; wan22_video is the original path, unchanged.
A workflow saved earlier loads with the default until the input is set.
Also included: the Alice, Ideogram Magic Prompt and prompt-database nodes that
were already registered in __init__.py but had not been committed.
Fixes: v1 {animal} / {label:a|b} wildcards were inert in the video nodes;
Image-to-Video save_to_file raised NameError; T2I padded short-prompt models.
.gitattributes added (text=auto) so CRLF working files stop showing as
whole-file changes. prompt_database.db and the cached Ideogram prompt are
git-ignored.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019VpCi6nw5UECMdZ7ayYJvN
44 lines
3.1 KiB
YAML
44 lines
3.1 KiB
YAML
id: illustrious
|
|
name: "Illustrious XL"
|
|
developer: "OnomaAI Research"
|
|
summary: "SDXL anime fine-tune on Danbooru. A tag model: tags are written with spaces, comma separated, and the quality tags come LAST. From v1.0 short plain-English phrases may be mixed in, but Onoma's own card says tags are more accurate."
|
|
modality: image
|
|
tasks: [t2i]
|
|
encoder: "SDXL CLIPs (77-token chunks)"
|
|
encoder_family: clip
|
|
caption_style: tags
|
|
validator: tags
|
|
languages: [en]
|
|
target_words: [25, 80] # about 15-45 tags. The earlier ~150 target only produced padding (same finding as NoobAI); Onoma states no length
|
|
negative: "yes"
|
|
negative_default: "worst quality, bad quality, low quality, lowres, bad anatomy, bad hands, extra digits, jpeg artifacts, signature, watermark, username"
|
|
negative_note: "Onoma's v0.1 card: Euler a, 20-28 steps, CFG 5-7.5. 'worst quality' and 'bad quality' are real training labels for this model, so they work in the negative. Onoma publishes no official negative string; this one is those labels plus common Danbooru negatives."
|
|
quality_tokens: required
|
|
suffix: "masterpiece, best quality"
|
|
quality_token_list: [masterpiece, best quality, good quality, absurdres]
|
|
quoted_text: none
|
|
weights: true
|
|
structure: "person count (1girl, 2boys, no humans), character name, series, rating if wanted (general, sensitive), general tags (appearance, clothing, pose, expression, composition, background, lighting), artist, then the quality tags, then an optional year tag."
|
|
rules:
|
|
- "Danbooru tags written with SPACES, never underscores: 'looking at viewer', 'long hair', 'hatsune miku'. Comma separated."
|
|
- "Follow the order in the structure. It is the order of the training captions; quality tags near the END, not at the start."
|
|
- "A tag that contains parentheses is written with them escaped, because bare parentheses are weight syntax in ComfyUI: 'ms paint \\(medium\\)', 'saber \\(fate\\)'."
|
|
- "Use one framing tag that fits (portrait, upper body, cowboy shot, full body); never two that conflict."
|
|
- "v1.0 and later: a short plain-English phrase may be mixed in for something no tag covers. Tags first."
|
|
avoid:
|
|
- "Underscores inside tags"
|
|
- "Sentences and narrative"
|
|
- "Photography and 3D-render vocabulary"
|
|
- "Quality tags at the start of the prompt"
|
|
- "Padding with near-synonyms to reach a length"
|
|
legacy:
|
|
prompt_style: booru_detailed
|
|
supports: [booru_tags, character_details, quality_emphasis, anime_specific]
|
|
length_guidance: "About 15-45 tags. Stop when the picture is described; do not pad."
|
|
sources:
|
|
- {url: "https://arxiv.org/html/2409.19946v1", supports: "Onoma's paper: literal prompts with spaced tags, training caption order (person count, character, rating, general, artist, quality, year), quality-tag percentiles", confidence: high}
|
|
- {url: "https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0", supports: "sampler, steps, CFG, framing-tag advice", confidence: medium}
|
|
- {url: "https://huggingface.co/OnomaAIResearch/Illustrious-XL-v1.0", supports: "natural language plus tags from v1.0, 'tag-based approaches are more concise and accurate', 1536 px native", confidence: medium}
|
|
checked: "2026-09-21"
|
|
confidence: high
|