✦ Smart Scene Generator

Think of this as IllustriousScenesPlus after it went through a growth spurt. Same scene-generation core - category, complexity, the Include * toggle grid for weather/props/pose/clothing - but IllustriousSmartSceneGenerator bolts on an expression picker, a full quality-tag block, a negative-prompt generator, and subject/upscale-tag fallbacks, so it can hand you a complete, ready-to-encode positive-and-negative pair instead of just a scene fragment. It's the "give me the whole thing" version.

What's layered on top of the scene core

The Emotion/Expression, Emotion Group, Emotion Weight, and Wrap Emotion fields fold in everything the standalone IllustriousEmotions node does, right into the scene pipeline. Then comes the quality-block system the README calls out as its v2.6 addition: Use Quality Block (default on) auto-prepends a token list - default masterpiece, best quality, ultra-detailed, highres - unless one's already present in your prompt. Quality Weight and Quality Weight Scheme (auto/comfy/A1111) decide whether that block gets grouped weighting like (masterpiece, best quality:1.2) or expanded per-token like (masterpiece:1.2), (best quality:1.2); Per-Token Quality Weights forces the per-token form even under comfy/auto.

Subject Fallback (none/1girl/1boy/solo/1girl, solo) injects a baseline subject tag when you've left Person Description off, with Subject Fallback Order choosing whether it lands before or after Upscale Tags (none/anime_hd/ultra_hd/photo_8k), a set of extra resolution/detail tokens. Negative Preset (standard/aggressive/anime_clean/photoreal/custom) generates a matching negative prompt; flip Generate Negative Output on to actually get it on the second output port, and use Custom Negative Preset to supply your own string when the preset is custom. Weight Interpretation and Token Normalization are worth reading as hints, not enforcement - the tooltips are explicit that the actual weighting happens at your encoding node, this just tags what convention the output was written for.

The same Enable TIPO Optimization block from IllustriousScenesPlus (TIPO Candidates/Flavor/Max Length/Seed) is here too, generating and ranking prompt variants before the quality block and fallbacks get layered on.

Outputs

Three of them: prompt, negative (empty unless Generate Negative Output is on), and metadata - a JSON string documenting what the node actually decided, including pruning info when the token budget forced it to drop upscale tags. Check metadata before assuming the model is ignoring part of your prompt; it'll often show you it pruned something you didn't notice.

Installing it

ComfyUI Manager: search "EasyIllustrious," install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/regiellis/ComfyUI-EasyIllustrious, then pip install -r ComfyUI-EasyIllustrious/requirements.txt inside your ComfyUI environment, then restart.

Where people get surprised

With this many toggles defaulting to on, a first run through this node tends to produce a noticeably longer, more heavily-tagged prompt than a beginner expects - the quality block, subject fallback, and TIPO optimization can all fire at once without you touching a single field. This is exactly the kind of behavior that confused a user on the pack's Reddit launch thread, who found random tags appearing in their prompt and traced it back to having a scene generator wired into their chain and quietly injecting content. If you want tight control, start by turning off Use Quality Block, Enable TIPO Optimization, and Generate Negative Output, confirm the bare scene output looks right, then re-enable pieces one at a time.