You want high-quality anime images. Sharp outlines, vibrant colors, accurate character details not the blurry, generic output you get from models that weren't built for this style.
Describe your scene. Get finished anime art.
Run it now on Floyo!
Why NetaYume Lumina
Base Lumina Image 2.0 is a strong general-purpose model. NetaYume is a fine-tune built specifically for anime. The difference shows immediately sharper outlines, more accurate anatomy, better clothing and accessory detail, coherent backgrounds.
It understands anime-specific conventions that general models miss: outfit layering, hair physics, accessory placement, expression nuance. You describe a scene and get something that looks like it belongs in a finished production, not a test render.
sharp outlines and vibrant colors by default
accurate character rendering clothing, accessories, hairstyles
coherent background composition
artist style steering via @ tags
best results at 1024x1024, 30 steps
How Prompting Works
NetaYume uses a fixed prefix. You don't touch it. Everything after the <Prompt Start> tag is your scene description.
Write what you want to see: character, setting, lighting, mood, clothing, pose. The model handles the rest.
Want a specific art style? Add @artistname in your prompt. The model supports artist tag steering browse the community reference sheet linked in the workflow notes for available tags.
Example prompt (after the prefix tag): "1girl, silver hair, long braids, shrine maiden outfit, standing in a forest clearing, soft evening light, detailed fabric, @artistname"
Key Settings
CFG : Default 4. Drop to 3 for looser, more creative output. Raise to 5 for tighter prompt adherence. Above 6 introduces artifacts without quality gains.
Steps: Default 30. Clean, detailed output at this setting. Going below 20 loses fine detail. Above 35 rarely improves anything.
Resolution: 1024x1024 is the sweet spot. The model was trained here. You can push to 1024x1536 for portrait compositions, but square is the most reliable.
Sampler Shift: Default 4. Leave it unless your outputs look overcooked or washed out. Drop to 2 or 3 if that happens.
Negative Prompt: Pre-filled with standard quality filters. Leave it as-is for most generations. Add specific artifacts you're seeing if results need correction.
What This Is Great For
Character design: The model handles anime anatomy, outfit layering, and accessory detail better than general-purpose models. Strong for character sheets and turnarounds.
Scene illustration: Backgrounds stay coherent, lighting reads as intentional, compositions feel finished rather than rough. Good for key art and scene concepts.
Style exploration: Artist tag steering lets you push the output toward specific aesthetic directions without LoRAs or additional fine-tuning.
Animation and storyboard concepts: Fast enough for rapid scene ideation, detailed enough to use as actual reference material.
What to Watch Out For
This is an anime-first model. It will not produce photographic realism. If you need photorealistic output, use Flux or SD 3.5 workflows instead.
Artist tags only work if the artist is in the model's training data. Unknown @ tags are ignored rather than causing errors you just won't see the style steer. Check the community reference sheet before relying on a specific tag.
Very complex multi-character scenes can lose accuracy. The model handles single and dual character compositions well. Three or more characters in one scene gets harder.
Going above CFG 6 consistently introduces color artifacts and over-saturation. Stay between 3 and 5 for clean output.