Video and image updates 2026-08-16 (Patreon)
Content
Just uploaded 70+ videos and 90+ images to Mega. I took a bit of a different approach to this update, mainly by including most of the raw footage that I made for the recent Lost Cat video (including lots of unused footage from both sequences), as well as a bunch of concept images for the werewolf character (who remains, annoyingly enough, unnamed), and a big spoiler-y folder full of images for the upcoming Lisette video - skip that if you want to go in blind but I thought it might be interesting to see and hear a bit about the process I'm testing out for it.
I also added a few older videos with sound added using SD 2.5 - I'll likely revert back to using Kling to add sound to older clips (mainly due to cost) but SD 2.5 is pretty great at it TBH.
And as usual in addition to the videos that I thought were worthy of keeping, I've included some interesting and (hopefully) entertaining failures.
As always, the links and details are over on Discord in the #content-updates channel. This content will be added to the main archive after a couple of weeks. If you'd rather not go to Discord, here's the direct link
As usual I'll move these over to the main archive folders in a few weeks.
The SPOILERIFFIC folder contains a bunch of images generated for the upcoming clip with Lisette and the big floofy guy, and is sort of somewhere between a moodboard and concept art. The idea is to think through the flow of the scene really loosely and then generate some images for each key scene beat. Then I can piece together a sort of storyboard sequence out of the ones that seem most promising, then refine those to make them more visually consistent, generate variants if needed, and then use the final results as initial images for I2V generations. Each of the images is made using Seedream 5, using Stable Diffusion images as references for compositing and editing.
As you can see, there's a fair amount of inconsistency with these images: character details aren't consistent, size relationships change, the scene/background/lighting is all over the place - it's a mess. But to be honest working entirely with Stable Diffusion to generate the seed images gives you the same issues, just worse.
Hence the refinement process I mention above: once I nail down a few images that mor eor less cover the main scene beats, I can then use editing tools to try to make them all a bit more visually similar, make sure that the characters don't change as much over time, etc. I'm also experimenting with AtlasCloud's "References" feature to define specific characters - so far it seems to do a good job of ensuring character consistency throughout a scene, but it may only work in Kling. For SD videos I'll be relying on multiple reference images.
Thanks again for the support - pretty much everything in this update is entirely due to the over 100 of you that have subbed here since launch. It's still crazy to me that people are willing to do that, but it's appreciated more than you know.