Nova Canvas cfgScale 1.1 to 10 vs Sume: no guidance field

cfgScale tells Nova Canvas how strictly to follow the prompt. Amazon's docs (read 2026-10-06) give a range of 1.1 to 10 with a default of 6.5. If your tuned settings include a cfgScale, you will not find the knob on Sume.

What happens if you send it

Sume's Image API checks every request against the model's catalog row. A parameter the row does not list returns 400 unsupported_parameter, and the API does not drop it silently. seed, output_compression and stream fail the same way, and provider.options must be empty. The port checklist lists the usual ones.

Ways to get tighter prompt adherence on Sume

• Change quality where a model lists it: Ideogram 4.5 takes low, medium or high; GPT Image 2.5 adds xhigh and max.
• Pick a model that follows text well for your case. FLUX.2 flex has steps and guidance on BFL's side but no passthrough on Sume.
• Write the must-keep parts in the prompt, and split a long instruction into one change per call.

List what a model accepts

Ask the catalog instead of guessing. This prints the parameter names a model lists, so you can strip the rest before sending.

import os, requests

model = "ideogram/ideogram-v4.5"
r = requests.get(
    f"https://api.sume.com/v1/images/models/{model}/endpoints",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    timeout=30,
)
r.raise_for_status()
for ep in r.json()["endpoints"]:
    print(sorted(ep["supported_parameters"]))