Some ai/site news that have been piling up. We did not get an official update yesterday, so here is some older and some newer stuff.
Update on image feeds (16th August 22:00 UTC): Still broken. Daily test image I published right now still shows "not published" or does not show at all: https://civitai.red/images/139830592
Image feeds: https://civitai.red/articles/33829/image-feed-out-of-sync
As much as I can tell the problem with image feeds I reported two three days ago has not been solved yet. All test images I posted were marked as "not published" instead of "scheduled" before publishing and it seems like they were not visible on the feed when being published (no initial reactions, only got some later on).
Kohya onsite trainer disfunctional: https://civitai.red/issues?id=60
There is an issue with the onsite trainer right now letting all Illustrious, Chroma and Flux Dev LoRA training runs using kohya engine fail.
New Anima 2.9B model: https://civitai.red/models/2855007/anima-29b?modelVersionId=3224434
So someone was mad enough to frankenstein the Anima model and add additional layers. The new model has now 40 instead of 28 layers and has 2.9 billion parameters instead of 2 billion. This is an effort to make Anima more robust to finetuning and adding more style and character knowledge. The model was already trained with additional 1M booru images compared to the original Anima with dataset cutoff July 2026. It is supported in Forge Neo and ComfyUI with latest commits already. Old LoRAs (trained with Anima 2B) still work depending on concept/style and how they were trained (text encoder trained LoRAs are probably doing better).
Edit: Forgot to mention it before, but the cover image was generated with Anima 2.9B and Kuro anima LoRA as well as Kyhustyle anima LoRA, trained with anima 2B base and preview3 respectively .
The creator plans to train on another 10M images from different sources to improve the knowledge of the model. This is a self-financed project. You can find support links on the projects huggingface page:
https://huggingface.co/Gazingstars123/Anima-2.9B
In-product feedback: https://github.com/civitai/civitai/pull/3901
So seems like they want to implement feedback options for different parts of the site to lower the amount of support tickets and it's starting with the image/video feed. Haven't seen this option yet.
Links to comments: https://github.com/civitai/civitai/pull/3786
You can now create a link to a comment now by clicking on the option tab of a comment. Here is a link to a random comment of mine: https://civitai.red/articles/33917/weekly-updates?highlight=2302954
This is something I missed on civitai before.
Collaborative Collections: https://github.com/civitai/civitai/pull/3719
A Membership feature. If you are a member you can open a collection for accepting submissions that get saved in a review queue. Also you can invite other users as Contributor (VIEW+ADD) or Manager (VIEW+ADD+MANAGE). Don't know how this works in detail, but it seems to be implmented already. "Editing Collection" dialogue now shows a message "Opening a collection to submissions is a member feature. Upgrade to let other people add to this collection." and a greyed out dropdown "Who can add to this collection".
Article thread at full depth: https://github.com/civitai/civitai/pull/3812
Maybe you noticed already that comment threads on articles expand to their deepest depth by default, so that you can see every comment without having to hit "Show 1 more" for every new comment. This goes to a depth of 10 comments as much as I can tell.
Profile on .com: https://github.com/civitai/civitai/pull/3794
You can let your profile on .com show a different cover image, bio and announcement now. This is done via "Customize Profile".
Remix gallery: https://github.com/civitai/civitai/pull/3810
As announced in a previous official article every creator gets a remix gallery now, where other creators can pay you buzz to display their remixes of your images. This gallery sits in your /images tab and there are new filter options for "Originals Only" and "Remixes Only".
Edit: Apparently the filter options always have been there (see the comment by IcebergMM) and I just didn't pay attantion, so seems the remix gallery isn't available at all right now.
The submitted images do not even have to be remixes apparently. Currently it does not seems to work, both filters only show me some images I posted to the model page of a model that got blocked because of POI. This is normal behavior for a not working image feed it seems. I don't think this will help to solve the issue of visibility, but who knows.
Featured images: https://github.com/civitai/civitai/pull/3855
New method how Featured Images are drawn from the featured collections pool. Don't looked into the details, read the PR yourself, if you are interested.
Creator Card on hover: https://github.com/civitai/civitai/pull/3846
Hovering over avatars, usernames and @mentions now reveals the creator card of a user.
Article Scan:
About three weeks ago the article scanner failed to scan cover images on multiple articles. I had to delay the publishing of this article by a day because of that. This also happened to an article by JustMaier and it was not visible for anybody on publish. So probably because of this article cover images get now scanned together with all other images and get requeued if the scan fails: https://github.com/civitai/civitai/pull/3360
Also other PRs were made to ensure the article owner gets informed properly when the article fails to get scanned:
https://github.com/civitai/civitai/pull/3359
https://github.com/civitai/civitai/pull/3338
That looks something like this:
Some other changes to article scanning were made:
https://github.com/civitai/civitai/pull/3314
https://github.com/civitai/civitai/pull/3336
There are probably more news that I forgot to mention, but the article is already long enough I think. I'll end this by introducing you to the work of ferretduck, who is developing tools for image generation/model creation. Unfortunately one of their articles describing the tool TensorPencil got stuck in the rescan and barely anyone saw it. Take a look here. If you want to craft images with support by an LLM, this is a nice tool to check out. Personally I tried ggufy another tool by ferretduck for quantizing base models. It works really fast quantizing a 2B model like Anima in seconds on CPU and it ships with an easy to use UI. Give it a try if you want to quantize a model.