babysea-team/sherin-avatar 🖼️🔢❓📝✓ → 🖼️

▶️ 64 runs 📅 Dec 2025 ⚙️ Cog 0.13.6 ⚖️ License

About

Ultra-fast image generation model optimized for creating high-quality avatars.

Example Output

Prompt:

"SHERIN, anime style, hentai style, flat color, cel shaded, 2D, hard lines, drawn street background, sharp focus, no bokeh, no blur, 1girl, dark-teal short hair, teal eyes, medium breasts, cleavage, casual clothes, smile"

Output

Performance Metrics

0.78s Prediction Time

0.79s Total Time

All Input Parameters

{
  "model": "schnell",
  "prompt": "SHERIN, anime style, hentai style, flat color, cel shaded, 2D, hard lines, drawn street background, sharp focus, no bokeh, no blur, 1girl, dark-teal short hair, teal eyes, medium breasts, cleavage, casual clothes, smile",
  "go_fast": true,
  "extra_lora": "https://huggingface.co/XLabs-AI/flux-lora-collection/resolve/main/anime_lora.safetensors",
  "lora_scale": 1.3,
  "megapixels": "1",
  "num_outputs": 1,
  "aspect_ratio": "1:1",
  "output_format": "png",
  "guidance_scale": 3,
  "output_quality": 90,
  "prompt_strength": 0.8,
  "extra_lora_scale": 1.25,
  "num_inference_steps": 4
}

Input Parameters

mask Type: string: Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored.
seed Type: integer: Random seed. Set for reproducible generation
image Type: string: Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored.
model Default: dev: Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps.
width Type: integerRange: 256 - 1440: Width of generated image. Only works if `aspect_ratio` is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation
height Type: integerRange: 256 - 1440: Height of generated image. Only works if `aspect_ratio` is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation
prompt (required) Type: string: Prompt for generated image. If you include the `trigger_word` used in the training process you are more likely to activate the trained object, style, or concept in the resulting image.
go_fast Type: booleanDefault: false: Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16
extra_lora Type: string: Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars'
lora_scale Type: numberDefault: 1Range: -1 - 3: Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora.
megapixels Default: 1: Approximate number of megapixels for generated image
num_outputs Type: integerDefault: 1Range: 1 - 4: Number of outputs to generate
aspect_ratio Default: 1:1: Aspect ratio for the generated image. If custom is selected, uses height and width below & will run in bf16 mode
output_format Default: webp: Format of the output images
guidance_scale Type: numberDefault: 3Range: 0 - 10: Guidance scale for the diffusion process. Lower values can give more realistic images. Good values to try are 2, 2.5, 3 and 3.5
output_quality Type: integerDefault: 80Range: 0 - 100: Quality when saving the output images, from 0 to 100. 100 is best quality, 0 is lowest quality. Not relevant for .png outputs
prompt_strength Type: numberDefault: 0.8Range: 0 - 1: Prompt strength when using img2img. 1.0 corresponds to full destruction of information in image
extra_lora_scale Type: numberDefault: 1Range: -1 - 3: Determines how strongly the extra LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora.
replicate_weights Type: string: Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars'
num_inference_steps Type: integerDefault: 28Range: 1 - 50: Number of denoising steps. More steps can give more detailed images, but take longer.
disable_safety_checker Type: booleanDefault: false: Disable safety checker for generated images.

Output Schema

Output

Type: array • Items Type: string • Items Format: uri

Example Execution Logs

Lora https://replicate.delivery/xezq/rLjewwLwWaXiZifZeOd03uu0VopJwDkVebKXRx4ezlAMgrZuC/flux-lora.tar already loaded
Extra lora https://huggingface.co/XLabs-AI/flux-lora-collection/resolve/main/anime_lora.safetensors already loaded
running quantized prediction
Using seed: 2026694887
  0%|          | 0/4 [00:00<?, ?it/s]
 75%|███████▌  | 3/4 [00:00<00:00, 16.04it/s]
100%|██████████| 4/4 [00:00<00:00, 14.54it/s]
Total safe images: 1 out of 1

Version Details

Version ID: fe1f7f9986145ce1c8797c4b26d3e75d462f69221e0e7c75c0cc81c282a300cb
Version Created: December 14, 2025

Run on Replicate →