---
library_name: karume
pipeline_tag: text-to-image
base_model: circlestone-labs/Anima-Base-v1.0-Diffusers
base_model_relation: quantized
license: other
license_name: circlestone-labs-non-commercial-license
license_link: https://huggingface.co/circlestone-labs/Anima/blob/main/LICENSE.md
tags:
  - text-to-image
  - webgpu
---

# Anima Turbo — Karume

## What is this

A distribution that bakes **Anima Turbo LoRA v0.2** into [circlestone-labs/Anima-Base-v1.0-Diffusers](https://huggingface.co/circlestone-labs/Anima-Base-v1.0-Diffusers) and converts it into the WebGPU
inference runtime **Karume**'s container format (a single safetensors file = weights +
a graph JSON embedded in `__metadata__`). Runs as-is in the browser and in Deno.

- A few-step distillation (from the LoRA) tuned for **8 steps / guidance 1**.
- Not readable by diffusers (it's a different container with an embedded graph); the reader is a pipeline that implements `anima/1`.
- Exporter used for the conversion: `karume/0.1.0`. The distribution manifest is `karume.json` (`karume/1`).

## Baked-in LoRA

Folded into the weights — not distributed as a separate file.

- **Name**: Anima Turbo LoRA v0.2
- **Author**: circlestone_labs (same author as the base model)
- **Source**: https://civitai.com/models/2560840?modelVersionId=2979642
- **File**: `anima-turbo-lora-v0.2.safetensors`
- **sha256**: `1b55e40bdb1d0e5a78cb498f245fccfdaae97823265db957d2aabdcf4cd3caf1`

Permissions listed on the source page (as of retrieval):

- `allowNoCredit`: true
- `allowCommercialUse`: Image / RentCivit / Rent
- `allowDerivatives`: true
- `allowDifferentLicense`: true

## Files

| Key | Variant | Path | Size | sha256 |
| ---- | ------- | ---- | ------ | ------ |
| `text_encoder` | — | `text_encoder/model.safetensors` | 1.11 GiB (1,194,225,572 B) | `79dc23f2d45c8f3e…` |
| `text_conditioner` | — | `text_conditioner/model.safetensors` | 257.34 MiB (269,838,156 B) | `a704ba27c865cd4e…` |
| `transformer` | f16 | `transformer/model.f16.safetensors` | 3.64 GiB (3,913,665,620 B) | `57c8a08be56c6fea…` |
| `transformer` | i8 | `transformer/model.i8.safetensors` | 1.83 GiB (1,962,558,660 B) | `df3cc9b539f30670…` |
| `transformer.rope_base` | f16 / i8 | `transformer/rope_base.safetensors` | 64.42 KiB (65,968 B) | `42db9a3fc796c45f…` |
| `vae_decoder` | — | `vae_decoder/model.safetensors` | 48.37 MiB (50,720,688 B) | `b50b65a028a8d108…` |
| `tokenizer` | — | `tokenizer/qwen2-tokenizer.json` | 3.35 MiB (3,514,619 B) | `0a7d6057ac8a2fe4…` |
| `tokenizer_2` | — | `tokenizer_2/t5-tokenizer.json` | 1.04 MiB (1,093,419 B) | `f86dfe21b12a175a…` |

Only the first 16 hex digits of the sha256 are shown (the full value and `size` live in `karume.json` — verify against that at the fetch layer).
Variant labels use the runtime's **storage dtype vocabulary** (`f16` / `i8`), not the `fp16` spelling common elsewhere in the ecosystem.

## Presets

| Preset | Weights | Compute |
| ------ | ---- | ---- |
| `f16` | `transformer` = `f16` | — |
| `i8` | `transformer` = `i8` | — |
| `w8a8` | `transformer` = `i8` | `linearCompute` = `i8a8` |
| `w8a8-a8` | `transformer` = `i8` | `linearCompute` = `i8a8` / `attentionCompute` = `i8a8` |
| `w8a8-s16` (default) | `transformer` = `i8` | `linearCompute` = `i8a8` / `attentionCompute` = `i8a8` / `attentionScoreStorage` = `f16` |
| `f16-c16` | `transformer` = `f16` | `linearCompute` = `f16` / `attentionCompute` = `f16` / requires `shaderF16` |

If no preset is given, it runs as `w8a8-s16` (the distribution's recommended default).

## Usage

```ts
import { AnimaPipeline, encodePng } from "jsr:@karume/models";

// The preset defaults to w8a8-s16.
using pipeline = await AnimaPipeline.fromPretrained("hdae/anima-turbo");
const image = await pipeline.generate({
  prompt: "1girl, solo, long hair, blue eyes, school uniform, masterpiece",
  seed: 42,
});
const png = await encodePng(image.data, image.width, image.height);
await Deno.writeFile("anima.png", png);
```

Weights are fetched once and cached (verified against `karume.json`'s `size` / `sha256`).
You can also load from a local directory (`AnimaPipeline.fromAssets`).

## Defaults

Any knob not passed to `generate()` is filled in from the manifest's defaults.

- **steps**: 8
- **guidanceScale**: 1
- **resolution**: 1024 × 1024
- **negativePrompt**: `low quality, worst quality, blurry, bad anatomy, jpeg artifacts`

At guidance 1, the second CFG branch is skipped, so **the negative prompt is not used**
(it only takes effect once guidance is raised).