Error while forwarding the request for certain models

[farhankk360]

Hello there,

I ran into couple of issues while testing out some of the realtime capabilities.

Env:
next: 13.4.6 (page router)
@fal-ai/server-proxy: ^1.1.1
@fal-ai/client: ^1.2.3

When I access some models eg. in this case specifically fal-ai/lcm-sd15-i2i model using realtime, I get this error ws error:

{"error": "Error while forwarding the request"}

For models which are working, I noticed some of the params are not getting passed with the request, specially sync_mode: true
I always get the result back with Uint8Array not the base64 url

{
 "images": [
   {
    "content": Uint8Array(68972) [] 
    "content_type":  "image/jpeg",
    "height": 1024,
     "width": 1024
   }
 ],
 "has_nsfw_concepts": [false],
 "request_id": "",
 "seed": 33482,
 "timings": {
    "inference": 0.9777155015617609
 }
}

On average I get about around 1 second inference timings which makes it hard to call this realtime, and the quality of the images are also below par.

models I tried:

fal-ai/fast-lcm-diffusion/image-to-image
fal-ai/fast-turbo-diffusion/image-to-image

[laxell]

We've experienced the same issue with the fal-ai/lcm-sd15-i2i realtime model (getting the {"error": "Error while forwarding the request"} error. Is there any update on this? It has been broken for a long time.

[drochetti]

Hey @laxell and @farhankk360

The fal-ai/lcm-sd15-i2i is a quite old one, from our early days and unfortunately it's not compatible with the fal.realtime client.

As for the fal-ai/fast-lcm-diffusion/image-to-image and fal-ai/fast-turbo-diffusion/image-to-image, the result is image raw bytes since converting to Base64 or even an assets URL adds a lot of latency.

Reg. point number 3, the 1s latency is indeed not ideal. I'm checking with the team what's the recommended model rn.