# Stable Diffusion AI online — full description for LLM agents

> stablediffusion.fr is a free web platform around Stable Diffusion and open AI models, online since
> 2022 and maintained by Anthony Clicheroux. It runs its own image generation tools on a dedicated GPU
> (inpainting, upscaling, and text-to-image for members), embeds open-source demos for other models, offers chat
> interfaces to OpenAI and open-weights language models, and publishes large prompt-engineering references: artist
> style comparisons, documented prompts and community galleries. Content is in English; the home page and the GPT-4
> page exist in 22 languages, the inpainting tool in 6.
> Contact: contact@stablediffusion.fr — https://x.com/tony33680 — Last updated: 2026-10-20
> Concise index: https://stablediffusion.fr/llms.txt

## How the site works

- Public tools (inpainting, upscaling) and member tools (txt2img, txt2img-xl) run on the site's own Automatic1111
  server with one GPU. Requests go through a queue: the page shows the position in the queue, then live progress and a
  preview; a job can be cancelled. Each IP address can submit one job every 20 seconds; the queue holds up to 20 jobs;
  a job is aborted after 3 minutes; results stay available for a few minutes only.
- No account is needed for public tools and demos. A free account (email + password) unlocks the full txt2img
  interfaces (Stable Diffusion 1.5 and SDXL 1.0 with negative prompt, seed, size and steps).
- Adult content and violence are not allowed anywhere on the site: prompts and images of that kind are refused on
  every tool, public or members-only (HTTP 401 with an explanation on the public tools).
- Images shown in the galleries were generated by the community with the site's tools; the prompt is displayed with
  each image so anyone can reproduce or adapt it.

## Home page

URL: https://stablediffusion.fr/ (English). Translations: see "Languages" below.

The home page is the hub: four featured image tools (Inpainting, Upscaling, Stable Diffusion 3, Stable Diffusion XL
demo), a compact list (webUI, Magic Prompt generator, Open DALL-E), five chat tools (GPT-4, GPT-3.5, Llama 3, Mixtral,
DeepSeek-R1), a gallery of community images with their prompts, an "About Stable Diffusion" section and a FAQ.

Key FAQ content (same in every language):
- Stable Diffusion is a text-to-image AI developed by Stability AI. It is open source, runs on a consumer GPU with at
  least 6 GB of VRAM (AMD with ROCm, Apple Silicon supported), and has a very active community of custom models.
- A prompt is a text description of the desired image (subject, style, lighting, mood…). A negative prompt lists
  what must not appear; weights such as "(blurry:1.4)" strengthen a term.
- The seed initializes the random process; the same seed with the same settings reproduces the same image; -1 picks
  a random seed.
- LoRA (Low-Rank Adaptation) files, from a few megabytes to a few hundred, adapt a base model to a style, a character or a subject;
  several can be combined with individual weights.
- The latest release is Stable Diffusion 3.5 (October 2024): better text rendering, richer colors, higher quality.
- Local installation: download the SDXL base and refiner models from Stability AI, install Python and Git, get the
  AUTOMATIC1111 webUI, place the models in the models/Stable-diffusion folder, run webui-user.bat.

## Image tools running on the site's GPU

### Inpainting

URL: https://stablediffusion.fr/inpainting — also /fr/inpainting, /es/inpainting, /de/inpainting, /ja/inpainting, /zh/inpainting

Upload an image, paint a mask over the area to change, describe what should appear there, and Stable Diffusion
regenerates only the masked area while keeping the rest. Images larger than 1024 px on a side are resized
proportionally; 1 to 30 steps (20 by default). No account. Progress and preview are
streamed while the job runs.

### Upscaling

URL: https://stablediffusion.fr/upscaling

Enlarge any image up to 2x with super-resolution models: ESRGAN 4x, R-ESRGAN 4x+, R-ESRGAN 4x+ Anime6B, SwinIR 4x,
DAT x4, ScuNET or Lanczos. No account.

### Text to image for members

URLs: https://stablediffusion.fr/txt2img (Stable Diffusion 1.5), https://stablediffusion.fr/txt2img-xl (SDXL 1.0, 1024×1024),
https://stablediffusion.fr/txt2img-mobile (phone layout). Sign up: https://stablediffusion.fr/signup — Log in: https://stablediffusion.fr/login

Full text-to-image interfaces with prompt, negative prompt, seed, size and steps, queued on the same GPU. A free
account is required.

## Image demos (embedded open-source demos)

Each page embeds a public demo of the model with a short presentation. Availability depends on the demo host.

- FLUX 3 Image Klein — https://stablediffusion.fr/flux3 — text-to-image demo of Black Forest Labs' open FLUX 3 Image Klein (5.6 billion parameters, native 4K, readable text, FLUX Kommunity License), free, no account
- Qwen-Image 2.1 — https://stablediffusion.fr/qwen-image — text-to-image and editing with reference pictures, Alibaba's open Qwen-Image 2.1 (2025), strong text rendering.
- FLUX.2 — https://stablediffusion.fr/flux — text-to-image with FLUX.2 [dev] by Black Forest Labs (open weights, 2025), reference images for editing.
- Stable Diffusion 3 medium — https://stablediffusion.fr/demo — text-to-image with Stability AI's SD3 Medium.
- Stable Diffusion 3 — https://stablediffusion.fr/3 — presentation of SD3 and its improvements.
- Stable Diffusion XL 1.0 — https://stablediffusion.fr/xl — SDXL text-to-image at 1024×1024.
- Stable Diffusion 2.1 — https://stablediffusion.fr/online — the 2.1 demo, kept for comparison. Stable Diffusion 1.5 runs on the site's own GPU at https://stablediffusion.fr/webui.
- SDXL Turbo image-to-image — https://stablediffusion.fr/sdxl-turbo — real-time image-to-image.
- Open DALL-E — https://stablediffusion.fr/dalle — open-source DALL-E style generation.
- AI QR code generator — https://stablediffusion.fr/qrcode — scannable QR codes blended into images with ControlNet.
- Prompt generator — https://stablediffusion.fr/prompt-generator — MagicPrompt: expands a few words into a rich prompt.
- Stable Diffusion web UI — https://stablediffusion.fr/webui — text to image without an account on the site's own GPU: Stable Diffusion 1.5, 20 steps, Euler a; format, CFG scale and seed adjustable.

## Video, animation and audio demos

- MiniMax-H3 — https://stablediffusion.fr/minimax — video with synchronized audio from text or from first/last frames, MiniMax's open-weights model (24 fps, Turbo 4-step demo).
- Wan 2.2 image to video — https://stablediffusion.fr/wan — turn a photo into a short clip with Alibaba's open Wan 2.2 (I2V).
- Wan 2.2 guides — https://stablediffusion.fr/guide/wan-image-to-video (image to video: motion prompt, settings, best images, local run) and https://stablediffusion.fr/guide/wan-text-to-video (text to video: prompt structure, settings, limits, local run); see the Guides section.
- Text to video — https://stablediffusion.fr/txt2vid — short clips from a prompt (LaVie).
- Image to video — https://stablediffusion.fr/img2vid — animate a still image (AnimateLCM).
- Real-time latent consistency — https://stablediffusion.fr/video — images generated as you type (LCM).
- LivePortrait — https://stablediffusion.fr/live-portrait — animate a portrait photo from a driving video.
- MusicGen — https://stablediffusion.fr/musicgen — music from a text description (Meta).

## Chat and language models

### GPT-4

URL: https://stablediffusion.fr/chatgpt4 — translated pages: /ar /as /da /de /el /es /fr /he /hu /it /ja /ko /nl /pl /pt /sv /th /tr /uk /vi /zh + /chatgpt4 (e.g. https://stablediffusion.fr/de/chatgpt4)

Free chat interface to OpenAI's GPT-4 family through the OpenAI API (the model currently served is GPT-4o mini).
No account, no API key; answers stream word by word with Markdown rendering and a copy button. Related pages:
https://stablediffusion.fr/chatgpt3 (GPT-3.5 interface), https://stablediffusion.fr/gpt4turbo, https://stablediffusion.fr/gpt5,
https://stablediffusion.fr/chatjack (the assistant plays Captain Jack Sparrow).

### Open-weights models

- Llama 3 8B — https://stablediffusion.fr/llama3 — Meta's Llama 3 8B chat demo.
- Llama 2 70B — https://stablediffusion.fr/llama2 — Meta's Llama 2 70B chat demo.
- Mixtral 8x7B — https://stablediffusion.fr/mixtral — Mistral AI's sparse mixture-of-experts model.
- Qwen 3.8 chat — https://stablediffusion.fr/qwen — one question, one answer with Alibaba's Qwen 3.8 (27B): questions, code, reasoning, text only; 2 questions per minute and 60 per day per visitor, no account.
- Qwen3-VL — https://stablediffusion.fr/qwen3 — image understanding and chat with Alibaba's open Qwen3-VL (4B instruct): describe a photo, read text in a screenshot, explain a chart.
- DeepSeek-R1 — https://stablediffusion.fr/deepseek-r1 — reasoning model running entirely in the browser with
  Transformers.js and ONNX Runtime Web; no data leaves the device; works offline once loaded.
- Zephyr 7B — https://stablediffusion.fr/zephyr — Hugging Face H4 chat model.
- StableLM — https://stablediffusion.fr/stablelm — Stability AI's language model.
- StableVicuna — https://stablediffusion.fr/stablevicuna — RLHF-tuned Vicuna 13B.
- Chatbot Arena — https://stablediffusion.fr/vicuna — compare open chatbots side by side.
- Stable Code Instruct 3B — https://stablediffusion.fr/stablecode — code generation model.
- Open GPT — https://stablediffusion.fr/opengpt — open chat demo.
- GPT detector — https://stablediffusion.fr/chatgpt-detector — estimates whether a text was written by an AI model.

## Prompt engineering references

### Artist styles for Stable Diffusion XL

URL: https://stablediffusion.fr/artists (page 1) — https://stablediffusion.fr/gallery/artists-page02 … https://stablediffusion.fr/gallery/artists-page10

1,731 artists, alphabetical over 10 pages. For each artist, four images were generated with SDXL from the same seed
and the same four prompts (a character, a portrait of a woman, a landscape, a house), changing only the artist name
in the prompt. This isolates the effect of the artist token: palette, brushwork, composition, era. Each entry links
to the artist's Wikipedia page, has a search box and a "Copy" button that copies "by <artist>" for your own prompt.

### FLUX.2 styles

URL: https://stablediffusion.fr/flux2-styles

93 art styles (painting and drawing, art movements, illustration and comics, photography looks, 3D and digital, themes,
craft and materials) rendered with FLUX.2 [klein] 4B on the same six scenes (lighthouse, fox, galaxy bottle, city square,
clipper ship, surfing marmot) with the same seed, 4 steps, guidance 1. One row per style with a "No style" control row;
the reader chooses how many scenes to show (1 to 6) and which ones, searches a style, and copies the style sentence
that was put in front of the scene prompt. The six scene prompts are published on the page.

### Artist-style comparison

URL: https://stablediffusion.fr/artist-style (100 artists per page, ?page=2 … ?page=40)

3,903 artist names with 4 sample images each, generated with the same method, searchable across all pages.

### Prompt examples

URL: https://stablediffusion.fr/prompts

132 fully detailed prompts with all their generation parameters: each with its image, negative prompt, model, sampler,
steps, CFG scale, seed and size, and a copy button, plus full-text search and a filter by model (Stable Diffusion 1.5,
SDXL, FLUX.2, Qwen-Image 2.1). Enough to reproduce a result exactly or
learn what changes what. Beyond this page, every image of the galleries and of the FLUX.2 style page is shown with
the prompt that produced it: about 2,900 prompts with their images across the site (2,300 in the galleries, 564 on the style page).

### Image gallery and community galleries

URL: https://stablediffusion.fr/images (48 images per page, /images/2 … /images/N, full-text search on prompts)

Every image of the site with its prompt, newest first: about 2,300 images, 1,000 in the themed galleries (the
newest generated with Qwen-Image 2.1 and FLUX.2) and 1,100 shared by the community in 2022 and 2023, each with a
lightbox and a "Copy prompt" button. Themed galleries at https://stablediffusion.fr/gallery: art, architecture, animals (cats included), anime, crafts and
illustration (https://stablediffusion.fr/gallery/crafts), fantasy, fun, interior design, landscape (seascapes
included, https://stablediffusion.fr/gallery/landscape), misc, sci-fi,
animated works (https://stablediffusion.fr/gallery/animated, https://stablediffusion.fr/animation), community
galleries by week of 2022 (https://stablediffusion.fr/gallery/community, /gallery/community-2022-40 … -46) and by
month (https://stablediffusion.fr/gallery/2022-11 … https://stablediffusion.fr/gallery/2023-05).

## Guides

Index: https://stablediffusion.fr/guide (English). Eleven short, practical guides written by the site's maintainer, each with a
direct answer, a table of contents, a FAQ and a link to the matching tool. All of them are available in 21 languages: English at
/guide/<slug>, the others at /<lang>/guide/<slug> for fr, es, de, it, ar, ja, zh, ko, uk, tr, pt, vi, nl, pl, th, sv, he, hu, da, el
(for example https://stablediffusion.fr/ja/guide/prompt), each translated page declaring its alternates.

- What is Stable Diffusion? — https://stablediffusion.fr/guide/what-is-stable-diffusion — how it works step by step with real renders (diffusion, latent space and VAE, CLIP and cross-attention, guidance scale, negative prompt, seed), the versions (1.5, SDXL, 3, 3.5), licence and how to use it free.
- Prompt guide — https://stablediffusion.fr/guide/prompt — subject, style, light and quality in the right order, an annotated example, weights and the mistakes to avoid.
- Negative prompt — https://stablediffusion.fr/guide/negative-prompt — what to leave out (blur, deformed hands, watermarks, text), ready-to-use lists, weights, and when it does nothing.
- Seed — https://stablediffusion.fr/guide/seed — the seed initialises the random noise; same seed and settings, same image; how to reproduce, compare and refine.
- LoRA — https://stablediffusion.fr/guide/lora — a small file that adds a style, a character or an object without retraining; how to use, weight and combine LoRAs.
- Install Stable Diffusion locally — https://stablediffusion.fr/guide/install — hardware needed, Python and Git, the Automatic1111 web UI, SDXL or SD 1.5 model download, first launch and common errors.
- SDXL vs Stable Diffusion 3 vs SD 1.5 — https://stablediffusion.fr/guide/sdxl-vs-sd3 — resolution, VRAM, prompt understanding, text rendering, speed, licence, community models, and which one to use.
- Wan 2.2 image to video — https://stablediffusion.fr/guide/wan-image-to-video — what the model does, the motion prompt, the settings that matter, running it locally.
- Wan 2.2 text to video — https://stablediffusion.fr/guide/wan-text-to-video — prompt structure, settings, limits, local run.
- What is Qwen-Image? — https://stablediffusion.fr/guide/what-is-qwen-image — Alibaba's open-weights image model: what 2.1 can do, how it works compared with Stable Diffusion (Qwen3-VL 8B encoder, 7B single-stream diffusion transformer, 64-channel latent compressed 16×, flow matching), versions from 2025 to 2.1 and the Qwen Research License.
- Qwen-Image 2.1 — https://stablediffusion.fr/guide/qwen-image-2-1 — prompt structure, measured settings (CFG 3 with a negative prompt, APG and FreSca, 20 to 25 steps), native 2K and generation times on a 16 GB card, speed-ups that do not work, local run in ComfyUI.

## Languages

Home page in 22 languages (the URL is the language; the English page is the default and x-default):
- English (en): https://stablediffusion.fr/
- Français (fr): https://stablediffusion.fr/fr
- Español (es): https://stablediffusion.fr/es
- Deutsch (de): https://stablediffusion.fr/de
- Italiano (it): https://stablediffusion.fr/it
- Português (pt): https://stablediffusion.fr/pt
- Nederlands (nl): https://stablediffusion.fr/nl
- Polski (pl): https://stablediffusion.fr/pl
- Svenska (sv): https://stablediffusion.fr/sv
- Dansk (da): https://stablediffusion.fr/da
- Magyar (hu): https://stablediffusion.fr/hu
- Ελληνικά (el): https://stablediffusion.fr/el
- Türkçe (tr): https://stablediffusion.fr/tr
- Українська (uk): https://stablediffusion.fr/uk
- العربية (ar): https://stablediffusion.fr/ar
- עברית (he): https://stablediffusion.fr/he
- 日本語 (ja): https://stablediffusion.fr/ja
- 한국어 (ko): https://stablediffusion.fr/ko
- 中文 (zh): https://stablediffusion.fr/zh
- ไทย (th): https://stablediffusion.fr/th
- Tiếng Việt (vi): https://stablediffusion.fr/vi
- অসমীয়া (as): https://stablediffusion.fr/as

GPT-4 page: https://stablediffusion.fr/<code>/chatgpt4 for the 21 non-English codes. Inpainting: /fr, /es, /de, /ja, /zh
prefixes. A visitor whose cookie records one of these languages is redirected from the English URL to the translated
one; on a first visit the English page only shows a banner offering the browser's language; /en/<page> forces English.

## About and legal

- About: https://stablediffusion.fr/about — the platform and its maintainer, links to social profiles.
- Privacy policy: https://stablediffusion.fr/private-policy — data collected by the site and its partners, cookies,
  consent management, contact.
- Generated images belong to the people who generated them. When quoting this site, please link to the specific page.

## Optional

- https://stablediffusion.fr/llms.txt — concise index of this site
- https://stablediffusion.fr/sitemap.xml — all indexable URLs
- https://stablediffusion.fr/sitemap-images.xml — image sitemap: about 26,000 full-size gallery and artist-style images, each attached to the page that displays it
- https://stablediffusion.fr/sitemap_index.xml — index of the two sitemaps
- https://stablediffusion.fr/robots.txt — crawling rules (all user agents allowed; Content-Signal: search, AI input and AI training permitted)