# DeepFloyd IF — card for AI agents

Open-source text-to-image diffusion model with a T5 text encoder and three cascaded stages, producing up to 1024x1024 px images.

Category: AI image generators · maker DeepFloyd Lab at StabilityAI · checked 2026-10-24

## Price
- Price: Free

## License
Modified MIT (code); separate DeepFloyd IF license for weights — https://github.com/deep-floyd/IF/blob/HEAD/LICENSE

- commercial: **conditions** (Commercial use with conditions)
  > to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software
  (https://github.com/deep-floyd/IF/blob/HEAD/LICENSE)
- attribution: **no** (No credit needed)
  > The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
  (https://github.com/deep-floyd/IF/blob/HEAD/LICENSE)
- in app: **conditions** (Ship inside your app with conditions)
  > to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software
  (https://github.com/deep-floyd/IF/blob/HEAD/LICENSE)

## Rules to follow
- Keep the copyright and license notice in all copies of the code you ship.
- Do not remove, disable or circumvent the inference filters (safety checker, watermarker) in the Software.
- Accept the DeepFloyd IF weights license on Hugging Face before downloading the model weights, and check LICENSE-MODEL for their terms.

## Can an AI agent use it by itself?
**Yes** · package: Agent installs a package or loads it from a CDN

## How to connect
- Package (npm/PyPI)
- PyPI: `pip install deepfloyd_if`
- Repository: https://github.com/deep-floyd/IF
- Website: https://github.com/deep-floyd/IF

## Notes
Added automatically. License read from the official pages.

## Numbers (AgentGid data)
- Gid Score: 16
- GitHub stars: 7,779
- Downloads last week: 4

## Lines to paste into your prompt or AGENTS.md

```
- DeepFloyd IF (AI image generators). Open-source text-to-image diffusion model with a T5 text encoder and three cascaded stages, producing up to 1024x1024 px images. Price: free; commercial use with conditions; no credit needed; ship inside your app with conditions.
  Connect: PyPI deepfloyd_if.
  Rule: Keep the copyright and license notice in all copies of the code you ship.
  Rule: Do not remove, disable or circumvent the inference filters (safety checker, watermarker) in the Software.
  Rule: Accept the DeepFloyd IF weights license on Hugging Face before downloading the model weights, and check LICENSE-MODEL for their terms.
  License: https://github.com/deep-floyd/IF/blob/HEAD/LICENSE (checked 2026-10-24; details: https://agentgid.com/assets/if/)
```

## Links
- AgentGid page: https://agentgid.com/assets/if/
- Is it free for commercial use? https://agentgid.com/assets/if/commercial-use/
- Source: https://github.com/deep-floyd/IF/blob/HEAD/LICENSE
- Source: https://github.com/deep-floyd/IF


## DeepFloyd IF: research text-to-image model, non-commercial

DeepFloyd IF is an AI model from DeepFloyd Lab and Stability AI that draws a picture from a text description, up to 1024 by 1024 pixels. It runs on your own computer with a large NVIDIA graphics card. The code is open, but the trained model may be used only for non-commercial research.

### Where to use it
- Pictures from a text description: Write what you want to see, and the model draws it in three steps, from a small picture up to 1024 by 1024 pixels. Research use only.
- Restyle an existing picture: Give it a picture and a style such as "origami" or "oil painting", and it redraws the picture in that style.
- Enlarge a small picture: Make a small picture four times larger with added detail, even if the model did not draw it.
- Repaint one part of a picture: Mark an area of a picture and describe what should be there, and the model fills only that area.
- Teach it a new subject: Train it on a few photos of one object or character so it can draw that subject in new scenes, for research.
- Compare image models in a study: Use it as a fixed, well-known model next to newer ones when you compare how models follow prompts. (idea, not from the official pages)

### How to start
- Sign in to Hugging Face and accept the license on the model page; the model cannot be downloaded before that.
- Install the Diffusers library and its helpers in a Python environment with PyTorch for your graphics card.

```
pip install diffusers accelerate transformers safetensors
```

- Log in from Python with your own access token, then run the three stages from the README example.

```
from huggingface_hub import login

login()
```


### License
The trained model is for non-commercial research only, and even its pictures may not be used for commercial purposes. The code has a modified MIT license that forbids turning off its safety filters.

### Good to know
- Not for paid products. The model license bans commercial and production use of the model and of the pictures it makes.
- You need a Hugging Face account, must accept the license on the model page, and must log in with an access token before the model downloads.
- It needs a large graphics card. The full set of models needs 24 GB of graphics memory; Diffusers can run it with about 14 GB.
- The safety checker and the watermark must stay on. The code license forbids removing or turning off these filters.
- The project looks finished rather than active. The model files date from June 2023 and the GitHub code was last changed in April 2024.

Checked 2026-11-12 against: https://github.com/deep-floyd/IF, https://github.com/deep-floyd/IF/blob/develop/README.md, https://github.com/deep-floyd/IF/blob/develop/LICENSE, https://github.com/deep-floyd/IF/blob/develop/LICENSE-MODEL, https://api.github.com/repos/deep-floyd/IF, https://pypi.org/pypi/deepfloyd-if/json, https://huggingface.co/DeepFloyd/IF-I-XL-v1.0, https://huggingface.co/api/models/DeepFloyd/IF-I-XL-v1.0, https://huggingface.co/docs/diffusers/api/pipelines/deepfloyd_if

### Mistakes to avoid
- Don't: Using IF pictures in a client's ad, product page or paid app → Do: Use it only for non-commercial research; pick a model whose license allows commercial use for production (The model license bans commercial and production use of the model and of data it produces.)
- Don't: Calling from_pretrained("DeepFloyd/IF-I-XL-v1.0") before the person has accepted the license and logged in → Do: Have the person accept the license on the model page, then log in with huggingface_hub login() or an HF_TOKEN (The weights are gated; the download fails without an accepted license and a token.)
- Don't: Keeping enable_xformers_memory_efficient_attention() on torch 2.0 or newer → Do: Delete those lines when torch is 2.0.0 or newer (The README says to delete all of them on newer torch.)
- Don't: Passing safety_checker=None or removing the watermarker to speed things up → Do: Keep the safety checker and watermarker, as in the README's stage 3 example (The code license forbids disabling the inference filters.)

### Questions
- Can I use the pictures in ads, a shop or a paid app? No. The model license forbids commercial or production use, and this also covers data the model produces.
- How big are the pictures? Stage 1 draws 64 by 64 pixels, and two more stages enlarge it to 256 and then 1024 pixels.
- Do I need an account? Yes. You need a Hugging Face account, accept the license on the model page and log in locally with a token.
- What graphics card do I need? 16 GB of graphics memory for the first two stages and 24 GB with the 1024-pixel stage; Diffusers can run it with about 14 GB.
- Must I credit DeepFloyd when I share it? Yes. When you pass on the model or works that include it, add a copy of the license and the attribution notice it gives.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
