TogetherAI image generation image_url input #2084
Description
Hi team, making this issue to keep track of our TogetherAI integration not accepting image_url
i don't think it's puterjs issue, but rather an upstream issue
here are different cases i found
google/flash-image-2.5
<html>
<body>
<script src="https://js.puter.com/v2/"></script>
<script>
puter.ai.txt2img("Draw an anime style version of this image.", {
model: "google/flash-image-2.5",
image_url: "https://assets.puter.site/doge.jpeg",
disable_safety_checker: true
}).then((image) => {
document.body.appendChild(image);
}).catch(e => console.error(e.error))
</script>
<img src="https://assets.puter.site/doge.jpeg">
</body>
</html>playground/:8 400 {"id":"oMZbc7k-2kFHot-9a78aa608b7f797c-PDX","error":{"message":"Unsupported use of 'image_url' parameter. This parameter is not supported for the selected model.","type":"invalid_request_error","param":null,"code":null}}
^ this should work, beecause img 2 img operation is in their recommended models
https://docs.together.ai/docs/recommended-models
(i know direct integration works, i'm just saying it doesn't with together ai)
black-forest-labs/FLUX.2-pro
<html>
<body>
<script src="https://js.puter.com/v2/"></script>
<script>
puter.ai.txt2img("Draw an anime style version of this image.", {
model: "black-forest-labs/FLUX.2-pro",
image_url: "https://assets.puter.site/doge.jpeg",
disable_safety_checker: true
}).then((image) => {
document.body.appendChild(image);
}).catch(e => console.error(e.error))
</script>
<img src="https://assets.puter.site/doge.jpeg">
</body>
</html>playground/:8 400 {"id":"oMZcM3F-2kFHot-9a78ade6def7a8fd-PDX","error":{"message":"An error has occurred please read documentation for additional details or contact support at support@runware.ai","type":"invalid_request_error","param":null,"code":null}}
^ this should also work, because they put this example code here accepting the image_url
https://www.together.ai/models/flux-2-pro
this is pretty frustrating given that newer models boast image editing (img2img) capabilities, but we can't utilize this because the input isn't working