Calling via API (Images / Video)

Standard OpenAI-style endpoints, authenticated by token. Images return base64 synchronously; video is an asynchronous task (create → poll → download).

Token Group

The token must belong to the media / media-gen group, otherwise you'll get a "No available channel" error. See Choosing a Group.

Images · Text-to-Image

POST https://byesu.com/v1/images/generations
Authorization: Bearer sk-your-token
Content-Type: application/json

Field	Required	Description
model	✅	Model name, see Image Generation
prompt	✅	Prompt
size		Width x height in pixels — determines both the aspect ratio and the resolution tier: long side ≈1024 → 1K, ≈2048 → 2K, ≈3840 → 4K. E.g. "1024x1024" (1:1 · 1K), "2048x1152" (16:9 · 2K), "3840x3840" (1:1 · 4K). Defaults to 1:1 at the base tier if omitted

curl https://byesu.com/v1/images/generations \
  -H "Authorization: Bearer sk-your-token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-2",
    "prompt": "A golden retriever running through a golden wheat field at dusk, cinematic",
    "size": "2048x1152"
  }'

Response (the image is a base64-encoded PNG):

{ "data": [ { "b64_json": "<base64 string>" } ] }

Images · Image-to-Image (Reference Images)

Use POST /v1/images/edits with multipart/form-data; reference images are uploaded as files (field name image, repeatable for multiple images, each ≤8MB; the maximum count varies by model, see Image Generation):

curl https://byesu.com/v1/images/edits \
  -H "Authorization: Bearer sk-your-token" \
  -F model="nano-banana-2" \
  -F prompt="Turn this photo into a watercolor painting" \
  -F size="1024x1024" \
  -F image=@ref1.png \
  -F image=@ref2.png

The response format is the same as text-to-image.

Video (Asynchronous, Sora-Style)

Three steps: create a task → poll its status → download the mp4.

import requests, time

BASE = "https://byesu.com"
H = {"Authorization": "Bearer sk-your-token"}

# 1. Create the task
task = requests.post(f"{BASE}/v1/videos", headers=H, json={
    "model": "gemini-veo31",
    "prompt": "Waves crashing against rocks, slow motion, cinematic",
    "seconds": "6",           # duration in seconds; available values vary by model
    "size": "1280x720",       # aspect ratio expressed as a size; 1920x1080 = 1080p (some models only)
}).json()

# 2. Poll until it finishes (tens of seconds to a few minutes)
while task["status"] not in ("completed", "failed"):
    time.sleep(5)
    task = requests.get(f"{BASE}/v1/videos/{task['id']}", headers=H).json()

# 3. Download the mp4
if task["status"] == "completed":
    mp4 = requests.get(f"{BASE}/v1/videos/{task['id']}/content", headers=H)
    open("out.mp4", "wb").write(mp4.content)

Image-to-video: create the task with multipart/form-data instead, putting the first-frame image in the input_reference file field (pass model / prompt / seconds / size along as form fields). runway-gen4-turbo requires a first-frame image.

Videos Are Billed Per Second

Video price = per-second unit price × seconds. Failed tasks are automatically refunded.

Errors

• "No available channel" → the token's group is wrong, see Choosing a Group
• too many reference images → you exceeded that model's reference-image limit
• 401 / insufficient balance / others → see Common Errors & Solutions