You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

SD1.5 Flow Matching Finetune (Evangelion 260K)

Model Details

This model is an experimental fine-tune of the NAI-v2 (Stable Diffusion 1.5) architecture, adapted to use a Rectified Flow / Flow Matching objective instead of the classical DDPM objective. The goal of this project was to test whether the Flow Matching objective accelerates the learning of new concepts and shifts the model distribution effectively to modern anime styles and characters (updating the base model's knowledge cutoff from Summer 2023).

Developed by: aipracticecafe
Objective: Rectified Flow / Flow Matching
Base Model: nai-anime-v2 (SD 1.5 based)
Resolution: Trained at 512px base resolution with aspect ratio bucketing. (Note: As training was strictly at 512px, the model's native capacity to generate 1MP images may be degraded compared to the base model). However it can generate 768x1024 images without problems.
Status: Experimental. This is the first model with somewhat stable results. The model has not been fine-tuned on a high-quality aesthetic dataset yet, so outputs might be unstable.

Prompt

1girl, amate yuzuriha, gundam gquuuuuux, beanie, black hat, blush, breasts, covered navel, gloves, green eyes, hat, indian style, (mecha pilot suit:1.1), medium breasts, normal suit \(gundam\), one eye closed, red hair, short hair, simple background, sitting, solo, pout, looking at viewers, closed mouth, white background, aqua eyes, bodysuit, (multicolored bodysuit:1.1), science fiction, v-shaped eyebrows, full body, dot nose, (anime coloring, virtual youtuber:0.5), general, (masterpiece, good score, absurdres, highres, very aesthetic, aesthetic, best quality, amazing quality:1.1), newest, new, year 2025, year 2024

Prompt

2girls, kaguya \(cho kaguya-hime!\), sakayori iroha, cho kaguya-hime!, animal ears, blonde hair, black hair, green eyes, clothing cutout, holding hands, crescent, crescent hair ornament, hair ornament, japanese clothes, kimono, long hair, looking at another, nail polish, open mouth, rabbit ears, rabbit girl, obi, detached sleeves, collarbone, red kimono, red nails, blue kimono, parted bangs, hood down, blue nails, parted lips, long sleeves, orange eyes, red eyeshadow, sash, multiple girls, facial mark, hood, yuri, medium hair, straight hair, , shoulder cutout, blue hoodie, belt, eye contact, interlocked fingers, forehead mark, dot nose, cowboy shot, (anime coloring, virtual youtuber:0.5), general, (masterpiece, good score, absurdres, highres, very aesthetic, aesthetic, best quality, amazing quality:1.1), newest, new, year 2025, year 2024

Prompt

1girl, souryuu asuka langley, neon genesis evangelion, rebuild of evangelion, evangelion: 3.0 you can \(not\) redo, long hair, blue eyes, orange hair, breasts, parted lips, official alternate costume, interface headset \(evangelion\), plugsuit \(evangelion\), plugsuit, pilot suit, red bodysuit, black eye patch, red gloves, interface headset, multicolored clothes, on throne, expressionless, one eye covered, headgear, skin tight, sitting on throne, crossed legs, head tilt, cowboy shot, throne, dominant, solo, sensitive, momoko \(momopoco\), toosaka asagi, ebifurya, solo, scenery, dot nose, (anime screencap, anime coloring, virtual youtuber:0.4), general, (masterpiece, good score, absurdres, highres, very aesthetic, aesthetic, best quality, amazing quality:1.05), newest, new, year 2025, year 2024

Dataset Details

Size: 260,000 samples.
Composition: Focused on relevant characters and artists to ensure adequate representation (e.g., >500 samples per character). Includes modern anime characters from series like Girls Band Cry, Oshi no Ko, Watashi ga Koibito ni Nareru Wake Naijan Murimuri!, Frieren, etc. Also it includes other series like Neon Genesis Evangelion and Date a Live.
Quality Clusters: The dataset was clustered into 4 quality buckets: worst score, bad score, good score, and masterpiece.
Prompting: Danbooru tags were upsampled using a SwinV2 tagger. Prompts follow a structured format with random shuffling of general tags every epoch. Loss was scaled using a tag weight float based on tag rarity to handle data imbalance.

Training Details

Hardware: 2x RTX 4090 (Took ~3 days)
Epochs: 20
Batch Size: 12 per GPU (Effective Batch Size: ~336 with 14 gradient accumulation steps across 2 GPUs). High batch size is crucial for learning new concepts as stated in the Illustrious paper.
Learning Rate: 2.5e-5 (Constant with 0.05 warmup)
Precision: Mixed Precision (fp32/bf16).
Optimizer: 8-bit AdamW (bitsandbytes)
Shift: 2
Latent Caching: VAE latents and TE tokens were precomputed and chunked.

Inference & Usage

Because this model uses a Flow Matching objective, it cannot be used with standard DDPM/DDIM samplers without modification.

WebUI: We have published an updated fork of the forge webui codeberg that adds native support for Rectified Flow models. ComfyUI: It should work using standard Flow Matching / SD3 sampling nodes (e.g., using ModelSamplingRectifiedFlow and a linear shift scheduler).

Tag Ordering

It's recommended to follow the structured prompt template:

1girl/1boy, character name, from what series, everything else in any order, artists, quality/metadata tags.

Including the artist after the series name can be useful for improving its effect.

Negative prompt:

(very displeasing, displeasing, bad score, worse score, bad quality,worst quality, worst detail:1.1),sketch, loli, child, patreon logo, watermark, signature, blurry, bad hands, bad anatomy, bad fingers, extra fingers, extra limbs, deformed limbs, very displeasing, displeasing, comic, speech bubble, lowres, twitter username, artist name, (light particles:0.4)

Recommended Parameters:

Sampler: Euler / DPM++ SDE (Must be Rectified Flow compatible)
Steps: 48
CFG Scale: 6
Shift: 3 - 4
Positive Prompt: masterpiece, good score, best quality, aesthetic, ...
Negative Prompt: worst quality, low quality, bad anatomy, bad score, ...
Resolution: 768 x 1024

Model Output Examples

Bias and Limitations

Unstable Outputs: Due to the lack of a final high-quality aesthetic fine-tuning phase, some outputs may be unstable or exhibit artifacts.
Resolution Degradation: Training exclusively on 512px aspect ratio buckets has reduced the model's ability to natively generate 1 Megapixel images (a feature present in the original NAI-v2).
Anatomy & Color: Standard biases from the Danbooru dataset apply.

License

This model is based upon Stable Diffusion 1.5. It is distributed under the terms of the CreativeML Open RAIL-M and the terms of the CC BY-NC-SA 4.0 license. This means that the terms of both licenses apply at the same time.

Downloads last month: -

Safetensors

Model size

0.9B params

Tensor type

F32