Skip to content
Images

Pokémon LoRA для Stable Diffusion 1.5

ранг 32 как финальный вариант

2 min readupdated

LoRA-адаптер для stable-diffusion-v1-5/stable-diffusion-v1-5, обученный на изображениях Pokémon с BLIP-подписями. Это один из двух запусков DATA 266 Lab 2 (rank 8 и rank 32 при одинаковых данных и шагах); автор выбрал rank 32 финальным, потому что его in-domain генерации ближе всего к тестовым изображениям по KID.

Ключевые особенности

  • Данные: reach-vb/pokemon-blip-captions, 833 изображения, после удаления дубликатов — 830; split 90/10 с seed 266: 747 тренировочных и 83 тестовых.
  • LoRA на attention-слоях UNet, rank 32, 6 377 472 обучаемых параметра.
  • Обучение: lr 1e-4 с cosine-расписанием, 3000 шагов, batch size 4 (около 16.1 эпохи), fp16, seed 266.
  • На одной RTX 4090: пик GPU-памяти 3.75 GB, 16.7 минуты.
  • Маленький и однородный датасет: персонажи на белом фоне; сложные промпты с несколькими объектами, количеством и цветами часто ломаются.

Пример

import torch
from diffusers import StableDiffusionPipeline
pipe = StableDiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
pipe.load_lora_weights("SawantTej/pokemon-sd15-lora-r32")
image = pipe("a cartoon bird with a hat on its head", num_inference_steps=30).images[0]
Подробнее Оценка: KID 0.0361 против 0.0947 у базовой модели; CLIP score 32.16 против 32.82; LPIPS между seed 0.755 против 0.830.
Обучение: одна RTX 4090, пик 3.75 GB, 16.7 минуты; 3000 шагов, batch 4, fp16.
Ограничения: возможен перенос стиля на посторонние промпты; BLIP-подписи шумные; KID посчитан всего на 20 изображениях; safety checker отметил 6 из 120 изображений, но это ложные срабатывания.
Только некоммерческие исследования и обучение: персонажи под копирайтом, данные cc-by-nc-sa-4.0, база — CreativeML OpenRAIL-M.
Discuss Telegram 0 18

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 3.2 GB · ready files5.0 GB
  • BF16weights 1.6 GB3.3 GB
  • FP8 / INT8recommendedweights 0.8 GB2.4 GB
  • 4-bitweights 0.5 GB2.0 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model stable-diffusion-v1-5.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # SawantTej/pokemon-sd15-lora-r32
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download stable-diffusion-v1-5/stable-diffusion-v1-5 --local-dir ~/models/stable-diffusion-v1-5
    hf download SawantTej/pokemon-sd15-lora-r32 --local-dir ~/models/pokemon-sd15-lora-r32
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to stable-diffusion-v1-5
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (3)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

stable-diffusion-v1-5/stable-diffusion-v1-5

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of stable-diffusion-v1-5 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

SawantTej/pokemon-sd15-lora-r32
Author
SawantTej
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments