ERNIE-Image Editing Model, August 2026: Another Missed Beta, and How the Community Is Coping

At midnight on July 31, 2026, the Reddit thread about the ERNIE-Image editing model got its latest reply: "Still waiting for it, maybe it will be out this week! Hopefully... Nope." Next to a screenshot of a "released by end of this month" promise from three months earlier, the comment section filled with resigned ellipses.

When ERNIE-Image went open-source in mid-April, the team shared a clear roadmap with the community: an editing-model Beta in July (ComfyUI nodes first), Diffusers integration plus an official API in August, and an 8-step Turbo editing version in September. July has now come and gone without a Beta. As of August 3, the GitHub repo still shows 21 commits — no editing-model activity whatsoever.

This isn't the first slip, and it won't be the last — but the community has already voted with its feet: stop waiting, and use what open source already has.

A Timeline of Waiting

Laid out flat, the arc is unmistakable: on April 15, the day of the open-source release, "when is the editing model coming" became the community's most-asked question. In early June, the team promised a July Beta, and a Reddit post titled "expected to be released by the end of this month" drew 334 upvotes and 51 comments. All through July, no official update. By early August, anticipation had curdled into jokes.

The sharpest comparison comes from a kindred spirit: Z-Image's editing model, Z-Image-Edit, was announced and then went silent for six months. As one comment put it, ERNIE is walking into Z-Image's holding pattern.

Community Sentiment

The top comments under that April Reddit post are basically a complete slice of community mood:

• "Still waiting for it, maybe it will be out this week! Hopefully... Nope."
• "WEN?"
• "They said the same for z-image edit... only to never be released."

Patience is wearing thin, but note one detail: nobody is questioning ERNIE-Image's value itself. The edit model is a nice-to-have, not a lifeline — generation capability is already good enough.

Official Signals

The only confirmed official facts are the roadmap itself (July Beta → August Diffusers + API → September Turbo) and the absence of a July release. Whether August's Diffusers/API milestone slides along with the Beta is unconfirmed. If a release does land in August, ComfyUI nodes and an API will most likely arrive together.

Qwen-Image-Edit: The Open-Source Leader

During ERNIE's silence, Qwen-Image-Edit has grown into the de facto open-source editing standard. Built on the 20B Qwen-Image, Apache 2.0, it leads all open models on the Arena image-editing leaderboard at 1241 Elo (the 2511 iteration sits at 1235).

The feature coverage is nearly complete: semantic editing (change actions or scenes while preserving character identity), appearance editing (backgrounds, clothing), in-image text editing, style transfer — even novel-view synthesis rotating 90° to 180°. The 2509 release adds multi-image editing across one to three images plus ControlNet conditioning (depth, edge, keypoint maps).

Other Open Alternatives

FLUX Kontext (12B) unifies generation and editing in one model, but its dev weights are non-commercial. Wan 2.6/2.7 Image (20B) is a dedicated editing pipeline, open and commercially usable. For ERNIE users, the smoothest interim path is still img2img + masking — community-verified for simple edits — and IP-Adapter for character consistency.

The Closed-Source Reference Frame

The top of the editing leaderboard is now dominated by closed models: GPT Image 2 (medium) at 1463 Elo; Microsoft's MAI-Image-2.5, released June 2, at 1401 (second place, and #3 on text-to-image), roughly $48 per 1,000 images; ByteDance's Seedream 5.0 Pro, which landed in July at 1240. These numbers say one thing: editing is the main battleground of the current image-model race, and whoever ships first owns the mindshare.

Workflow Advice for Right Now

Stop waiting and use the toolbox that exists:

1. Simple local edits: ERNIE-Image generation + img2img masked redraw
2. Character-consistent edits: ERNIE-Image + IP-Adapter
3. Professional editing: ERNIE-Image base image + Qwen-Image-Edit refinement
4. Text changes: use Qwen-Image-Edit's in-image text editing directly

All of this is orchestratable in ComfyUI today, and you can migrate smoothly once the official ERNIE editing model lands.

What to Expect in August

Per the roadmap, August was supposed to bring Diffusers integration and an official API. If the Beta slips into August, the most likely shape is ComfyUI nodes and the API launching together; the September Turbo edition (8-step inference) depends on Beta progress. Z-Image-Edit's ghost warns of "forever delayed" risk — but Baidu's track record on the generation side (open source in April, Turbo in May, ecosystem coverage within six months) is reason for cautious optimism.

Until then, Qwen-Image-Edit has already demonstrated, in open source, exactly what an editing model should look like. ERNIE's version is worth waiting for — but you don't have to wait idly.