Which providers give free image generation
Free image generation is less common than free chat. We checked each provider on its official page. The table shows what we could confirm.
| Provider | Models | Free allowance (official) | Notes |
|---|---|---|---|
| Cloudflare Workers AI | FLUX.1 schnell, FLUX.2 klein, FLUX.2 dev, Leonardo Phoenix and Lucid Origin | 10,000 neurons each day on the Free and Paid plans | FLUX.1 schnell uses 4.80 neurons for each 512x512 tile and 9.60 neurons for each step |
| NVIDIA (build.nvidia.com) | FLUX.1 dev and other hosted models | Trial service under the NVIDIA API Trial Terms of Service | FLUX.1 dev is for non-commercial use. NVIDIA does not publish a fixed image quota |
| Google AI Studio | Gemini 2.5 Flash Image | Not available on the free tier | Paid tier only, so the free pool cannot use it |
| Pollinations | Several image models | Not confirmed | FreeLLMAPI has an adapter. The official docs describe paid balance, and we could not confirm a free allowance today |
By our calculation, a FLUX.1 schnell image of four 512x512 tiles with 4 steps uses about 58 neurons. That gives about 170 images each day from the free Cloudflare budget. Chat, speech and embedding calls on Workers AI use the same budget.
Step 1: Install FreeLLMAPI and add keys
FreeLLMAPI is open source. It runs on your computer in Docker. Your provider keys stay on your computer.
Install (macOS or Linux)
curl -fsSL https://freellmapi.co/install.sh | bashOpen the dashboard at http://localhost:3001. Add a Cloudflare Workers AI key on the Keys page. The key has the form account_id:token. Then open Models, Image to see the image models and turn them on or off. Copy your unified key from the Keys page.
Step 2: Generate an image
The endpoint uses the OpenAI request format. Only prompt is required. n sets the number of images, from 1 to 4. size takes a value such as 1024x1024.
curl
curl http://localhost:3001/v1/images/generations \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "prompt": "a red bicycle in front of a blue door, photo", "size": "1024x1024"}' \
-o image.jsonThe response is JSON. Each item in data has a b64_json field with the image bytes in base64. The model and provider fields show which model made the image.
Python: generate and save the image
import base64
from openai import OpenAI
client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")
resp = client.images.generate(
model="auto",
prompt="a red bicycle in front of a blue door, photo",
size="1024x1024",
n=1,
)
raw = base64.b64decode(resp.data[0].b64_json)
ext = "png" if raw[:4] == b"\x89PNG" else "jpg" # providers return PNG or JPEG
with open(f"image.{ext}", "wb") as fh:
fh.write(raw)
print("Saved image." + ext)Write prompts that work on FLUX schnell
- Put the subject first, then the setting, then the style. For example: "a red bicycle, blue door, morning light, photo".
- Keep the prompt shorter than 2,048 characters. Cloudflare rejects longer prompts on FLUX.1 schnell.
- Do not ask for long text in the image. Small fast models often spell words wrong.
- FLUX.1 schnell on Cloudflare accepts only the prompt through this adapter. The
sizevalue has no effect on that model. - Send one image for each request when you test prompts. Each extra image uses more of the daily budget.
Limits and pitfalls
- Commercial use: the model license still applies on a free tier. FLUX.1 dev on NVIDIA is for non-commercial use. Read the license of each model before you publish images.
- Shared budget: the Cloudflare 10,000 neurons are for the account. Large images and more steps use more neurons.
- Draft small: Cloudflare counts cost in 512x512 tiles. A 1024x1024 image has four tiles, and a 512x512 image has one. On models that accept
size, make drafts at 512x512. Make the final image at full size. - Base64 only: plan to decode
b64_json. Do not expect a hosted image URL. - Different output: with
model: "auto", two requests can go to different models. Pin a model id when you need the same style each time. - Slow responses: image models take several seconds. Set a client timeout of 60 seconds or more.
/v1/videos/generations. Video jobs can take minutes and stop after a 5-minute timeout with error 504.Troubleshooting
| Problem | Cause | Fix |
|---|---|---|
Error 400 prompt is required | The body has no prompt | Add a prompt string |
| Error 429 or a budget error | The daily free budget is used | Wait for the daily reset, or add a key from another provider |
| Error 503 | No image model is enabled | Add a key and turn on a model on Models, Image |
| The image file does not open | The saved extension is wrong | Check the first bytes, as in the Python example |
| The size is not what you asked for | The model ignores size | Pin a model that accepts a width and a height |
Frequently asked questions
Is there a free image generation API?
Yes. Cloudflare Workers AI gives 10,000 neurons each day, which you can use for FLUX image models. NVIDIA hosts FLUX models under trial terms. FreeLLMAPI sends requests to these providers through one key.
How many free images can I make each day?
On Cloudflare, a FLUX.1 schnell image of four tiles with 4 steps uses about 58 neurons. That is about 170 images each day, if no other Workers AI model uses the budget.
Is Nano Banana free through the API?
No. Google lists Gemini 2.5 Flash Image as not available on the free tier of the Gemini API. Only the paid tier includes it.
Which image sizes can I request from the free image API?
Send size as width x height, for example 1024x1024. Some models ignore the size. FLUX.1 schnell on Cloudflare uses only the prompt through FreeLLMAPI.
Can I use free AI images commercially?
It depends on the model license and the provider terms. For example, FLUX.1 dev on NVIDIA is for non-commercial use. Read the license of each model.
Related guides
Sources
- FreeLLMAPI docs: Images, video and text-to-speech (checked 20 November 2026)
- Cloudflare Workers AI pricing (checked 20 November 2026)
- Cloudflare Workers AI: flux-1-schnell (checked 20 November 2026)
- NVIDIA build: FLUX.1 dev (checked 20 November 2026)
- Gemini Developer API pricing (checked 20 November 2026)
- Pollinations API docs (checked 20 November 2026)