Higgsfield's six launches in three days: what to re-test
Higgsfield's changelog lists six items from 2026-09-29 to 10-01. A re-test list for an ad team, with the Sume route to check for each one.

Re-test one thing per Higgsfield launch, not all six: a text-in-image prompt, a batch of ad stills, a restyle clip, and the agent hand-off. The Higgsfield changelog (read 2026-10-02) lists six entries between 2026-09-29 and 2026-10-01, and each maps to a check you can run against whatever tool you already use.
This post does not rank anything. It lists what the changelog says, then points to the Sume page that tells you whether the same job is possible there.
What did Higgsfield list, and what should I check?
The table keeps each claim to the changelog wording. The last column is a check, not a verdict.
| Date | Changelog entry | Check on Sume |
|---|---|---|
| 2026-10-01 | FLUX 3 Image: complex prompts, more accurate text, text-to-image and image-to-image up to native 4K | Look for the model id in GET /v1/images/models |
| 2026-09-30 | Ideogram 4.5: reference edits and text replacement that preserve layout, 1K or 2K | Same call; read input_references and resolution descriptors |
| 2026-09-30 | Ads Studio: website links to static ad creative, batch up to 20 ads per product | n on the Image API, or Format bulk runs |
| 2026-09-30 | Genjutsu Restyle: redraw a video in a new style, 20+ animation styles | higgsfield-genjutsu in the video catalog |
| 2026-09-30 | ChatGPT extension with /effects, /product and /genjutsu commands | Hosted MCP at https://mcp.sume.com/mcp |
| 2026-09-29 | OpenAI dots agents can generate on Higgsfield through a plugin | callback_url or webhook mode on a job |
How do I check a model id on Sume?
Do not guess ids. The Image API docs say each model publishes typed capability descriptors, and a request that sets a parameter the model does not list returns 400 unsupported_parameter. So list the catalog first, then read the endpoint record for the model you care about. If the id is not there, Sume does not serve it today.
Can Sume batch 20 ad stills like Ads Studio?
Two documented routes exist. The Image API takes n from 1 to 10 per call, with lower per-model ceilings, so 20 stills is at least two calls. Bulk runs queue up to 100 Format runs with a concurrency window. Neither one reads a website link and builds a brand setup for you; you pass the prompt and product image yourself.
What about the restyle and agent entries?
The video docs describe higgsfield-genjutsu as Motion Transfer: one source video plus 1 to 8 reference images, 480p or 720p, 4 to 30 seconds, listed only when its provider is configured. There is no list of named styles. For agent hosts, the MCP quickstart shows how to add the hosted server.
Run each check once, save the job ids and costs, and repeat in a month. A changelog line is a claim until your own call confirms it.
Sources
Related posts
More in Use cases
- Check 200 holiday clips before upload with a free probe-only inspect
A video-inspect call with frames false is a probe with no stills and no charge. Use it as a pre-upload audio gate, then pull stills only for clips you doubt.
- 300 holiday UGC clips on one plan: pace around queue_full 429s
A Pro workspace holds 4 processing and 20 queued jobs. Submit 300 avatar clips, treat queued as normal, and retry queue_full with the same idempotency key.
- Hour-long podcast video to clips: the 1800-second source cap
Sume's trim, detach and inspect tools read sources up to 1800 seconds, so a 60-minute episode must be split before import. Limits and a safe split plan.
- How many clips for a 10-minute faceless video? Timeline slot math
A 10-minute faceless video fits Timeline 1.0 comfortably: up to 200 slots, a render near $1.00, and limits on fades and single-pass renders to plan around.
Written by Sume