Images from a sentence, made in Switzerland.
Switch images on and ask: the model writes the prompt for FLUX.1 [schnell], an open model from Black Forest Labs (Germany). Attach photos of a face, and PhotoMaker V2 makes a new picture of that person. Both run at Infomaniak, in Switzerland. The image comes back to your browser — Evanesca keeps no copy of it — and you pay for the seconds it took: about €0.02 for a FLUX.1 [schnell] image, plus the model's tokens, as for any message.
Checked on 13 Oct 2026
Opening soon: what follows describes the service as it will work.
- Price per second
- €0.00924VAT included, per second of generation
- FLUX.1 [schnell]
- ≈ €0.02about 2 s per image
- PhotoMaker V2
- ≈ €0.05about 5 s per portrait
- Where
- SwitzerlandInfomaniak, outside the EU, adequacy decision
The two models
Two open-weight models, run by Infomaniak in its data centres in Switzerland.
| Model | What it does | Licence | Prompt | Typical time | Typical price |
|---|---|---|---|---|---|
| FLUX.1 [schnell]Black Forest Labs (Germany) | An image from a text | Apache-2.0 | 400 characters | 2 s | €0.0185 |
| PhotoMaker V2TencentARC | A new picture of a person, from 1 to 4 photos of their face | Apache-2.0 (SDXL base: CreativeML OpenRAIL++-M) | 350 characters | 5 s | €0.0462 |
How it works
- You switch images onImage generation is a switch in the chat, off by default. Once it is on, the model can make an image when you ask for one.
- You describe what you wantIn your own words. For a portrait, attach 1 to 4 photos of the face to your message.
- The model writes the promptIn English — at most 400 characters for FLUX.1 [schnell] and 350 characters for PhotoMaker V2, the lengths Infomaniak accepts — and picks a format: square, landscape or portrait.
- Infomaniak generates itIn Switzerland. One image per call, 4 at most per message.
- It comes back to youThe image shows under the answer and stays with the conversation, in your browser. The model doesn't see it: it only learns that the image was made.
Sizes and limits
- Sizes
- 1024 × 1024 (square), 1792 × 1024 (landscape) and 1024 × 1792 (portrait).
- Images
- One per call, 4 at most per message in the chat, 12 at most per task in Agent mode.
- Prompt
- At most 400 characters for FLUX.1 [schnell] and 350 characters for PhotoMaker V2.
- Face photos (PhotoMaker V2)
- 1 to 4, as JPEG or PNG.
- Agent mode
- FLUX.1 [schnell] only, and no images switch: the agent may make images for any task, within its cap.
What it costs
Infomaniak bills generation by the time it takes, and so does Evanesca: €0.00924 per second, VAT included. The seconds are measured by Evanesca's server around the call — network and queue included — and rounded up to the next second, one second at least. The message itself is billed as usual: the model's tokens, to write the prompt and to answer.
| What you ask for | Seconds | Price |
|---|---|---|
| One FLUX.1 [schnell] image | 2 | €0.0185 |
| 4 FLUX.1 [schnell] images, in one message | 8 | €0.0739 |
| One PhotoMaker V2 portrait | 5 | €0.0462 |
| 12 images, the most an Agent task can make | 24 | €0.2218 |
Typical times measured on 7 October 2026; a landscape or portrait image can take a little longer. These prices count the generation only: the model's tokens for the message come on top, and the exact cost shows under the answer. Every price
A face is personal data
- A photo of a face is personal data. Only send photos you have the right to use: your own, or those of a person who has agreed to it.
- The photos go to Infomaniak, in Switzerland, for the generation — and, like any image you attach, to the chat model you chose if it reads images — every model Evanesca serves runs in Europe. Their metadata (GPS position, device) is removed in your browser before upload. Evanesca keeps no copy of them.
- Never make, without their consent, a picture of a real person that could pass for real — and never an intimate or sexual image of a real person, even with their consent.
What is kept
- The image
- No copy at Evanesca: it comes back to your browser and stays there with the conversation, until you delete it.
- The prompt and the photos
- No copy at Evanesca: they pass through for the message, then are forgotten.
- Billing
- For each image: the model, the date and the cost.
- Infomaniak
- Infomaniak states that queries are “neither recorded nor used to train models” (Infomaniak AI Tools).
Marking generated images
The AI Act asks for images generated by AI to be marked in a machine-readable format, so that they can be detected as generated (art. 50(2)). At Evanesca, every generated image — in the chat, Agent mode and the API — carries IPTC metadata (XMP) declaring it made by a trained algorithm (“trainedAlgorithmicMedia”), with the description “AI-generated with Evanesca”. The pixels do not change. It is metadata, not a watermark: a screenshot or a re-encode can remove it. Where Evanesca stands on Article 50.
The licences
- FLUX.1 [schnell], by Black Forest Labs (Germany) — Apache-2.0. Model card
- PhotoMaker V2, by TencentARC — Apache-2.0 (SDXL base: CreativeML OpenRAIL++-M). It is built on SDXL, whose base model's OpenRAIL licence sets use restrictions. Model card
These licences govern the models. What you may do with an image also depends on what it shows: a trademark, an artwork or a person has rights of their own.
What is not allowed
- No picture of a real person without their consent, and no intimate, sexual or degrading image of a real person, even with it.
- Nothing sexual involving a minor, ever.
- No image made to deceive: a fake document, fake evidence, a fake photo of an event presented as real.
- No hateful or harassing image.
- Respect for others' rights: logos, characters and artworks belong to their owners.
The terms of use apply to every image.
Frequently asked questions
Where are the images made?
At Infomaniak, in Switzerland — outside the EU, under an EU adequacy decision. Both models, FLUX.1 [schnell] and PhotoMaker V2, run there.
How much does an image cost?
€0.00924 per second of generation, VAT included: about €0.02 for a FLUX.1 [schnell] image and €0.05 for a PhotoMaker V2 portrait — plus the model's tokens for the message, as for any answer. The exact cost shows under the answer.
Does Evanesca keep my images?
No copy: the image comes back to your browser and stays with the conversation until you delete it. Evanesca keeps the model, the date and the cost, for billing.
Can I make a portrait from my own photo?
Yes, with PhotoMaker V2: attach 1 to 4 photos of your face and describe the scene. Only use photos you have the right to use.
Are the images marked as AI-generated?
Yes: the images Evanesca generates — in the chat, Agent mode and the API — carry IPTC metadata marking them as AI-generated, the machine-readable marking the AI Act asks for (art. 50(2)). It is metadata, not a watermark: a screenshot or a re-encode can remove it.
Does the model see the image it made?
No: it only learns that an image was made, so it can neither describe nor fix it. Ask again with a new description.
Can the agent make images?
Yes: up to 12 per task, with FLUX.1 [schnell] — in Agent mode there is no images switch; each image is billed within the task's cap.