Anything V5

Anything V5

Generates anime-style images from text prompts, as a re-upload of the Anything V5 checkpoints, including the Ink and PRT variants. It suits artists who write precise prompts and people who train LoRAs.

Made by genai-archive on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Anything V5 is a text-to-image model in the Anything series of anime-style checkpoints. This repository is a re-upload under the CreativeML OpenRAIL-M license, and it holds the AnythingV5Ink and AnythingV5PrtRE variants.

The files came from ModelScope, not from Hugging Face, so some files from the original repository may be missing. The card says the file integrity was checked with a checksum.

Who it suits: Digital artists, hobbyists and LoRA trainers who work with anime-style images suit it best. The author says V5 needs precise prompts to look good.

Industry | Who uses it | How they use it | Example
Illustration | Digital artists | Make anime-style pictures from written prompts | Write a precise prompt for a character and scene
Hobby art | Web UI users | Compare the Anything versions side by side | Try the V5 Prt version against V3.2++
Model training | LoRA trainers | Use V5 as the base for LoRA training | Train a LoRA that works on most checkpoints

Anyone who wants quick results from a one-word prompt can skip it, since the author asks for precise prompts.

Where to use it: You run it yourself in Python with diffusers, or in a web UI that loads Stable Diffusion checkpoints.

• Windows, macOS or Linux with Python installed
• Python scripts that use the diffusers library
• A web UI with a model selector and an optional external VAE

Use case | Who | What they get back
Text to image | Artists | Anime-style images that follow your prompt
LoRA training | Trainers | A base checkpoint that fits most other checkpoints
Version comparison | Hobbyists | A feel for how V3.2++, V5 and Prt differ

How to set it up: You need Python and enough disk space for weights of about 3.4 GB in F32.

• Python on Windows, macOS or Linux
• The diffusers library, or a web UI that takes checkpoint files
• An external VAE if you pick the NoVAE version

1. Install Python.
2. Install the Hugging Face hub tools with pip.
3. Download the model files with the hf command.
4. Pick a version and load it in your tool of choice.

pip install -U huggingface_hub
hf download genai-archive/anything-v5

How to use it: The card shows no run commands, so you choose a version and load it as you would any Stable Diffusion style checkpoint. The author names V5 Prt as the most recommended version.

Tool | What it does
V5 Prt | A specially trimmed V5, which the author recommends most
RE labels | Fixed versions that address problems such as clip
V3.2++ Ink | Replaces the old V3, with weaker support for LoRAs built on NAI
NoVAE | Holds no VAE, so you pick an external VAE in the web UI

Example: You pick the V5 Prt version and write a precise prompt that names the character, the pose and the style you want. The author warns that V5 and later give good results only with precise wording, not with a bare one-word prompt.

For newcomers and experts: Think of this as a set of anime-style picture makers, with several versions to choose from. Your first small try is to download the files and generate one image with the V5 Prt version.

Under the hood, the series rests on five bases: V1, V2, V3, V3.2 and V5. Every other file is a variation of one of them. The author says A-Ink uses a base trained on many Niji images, with further training on Stable Diffusion images. LoRAs trained on V3.2++ work on few other checkpoints, so the author suggests V5 for training.

Pick a version, such as V5 Prt
Load it with diffusers or a web UI
Write a precise prompt
An anime-style image comes out

Good to know: This is a re-upload with its original license, and the author points out which versions are not his.

• Size: 860M parameters, about 3.4 GB in F32, stored as Safetensors
• Other versions: Hugging Face lists 0 quantized and 1 fine-tuned version made by other people
• License: creativeml-openrail-m, and the card quotes the grant in its Section II
• Cost to run: none named
• Data it reaches: the prompts you give it, and the model files downloaded from Hugging Face
• Not for: Anything 4.0 or 4.5, which the author says he did not make, or 2.5D styles, where he points to AOM2

On Hugging Face, 13 November 2026

• Task: text to image
• Library: diffusers
• Parameters: 860M
• Weights: about 3.4 GB in F32; about 430 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 12,112
• Likes: 33
• License: creativeml-openrail-m
• Fine-tunes: 1
• Tags: diffusers, safetensors

Links

• Model: https://huggingface.co/genai-archive/anything-v5
• Files: https://huggingface.co/genai-archive/anything-v5/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download genai-archive/anything-v5

Then follow the usage in its model card. Nothing in this kit runs by itself.