Home Creators Posts Import Register Favorites Logout

Content

Lately I've been experimenting with using more reference images in a bid to improve the continuity of the background scenes, overall visual style and tone, and character detail between the various clips that make up a longer scene.

In addition, I've started approaching scene planning from a bit more of a traditional storyboarding standpoint, with a goal of getting the clips that I need with less randomness, less wasted time and money.

So I've been taking my initial Stable Diffusion renders, then using them to generate a set of images in Nano Banana that show the setting from various angles (images like the ones at the top of this post). I cherry-pick the best of them, then use Nano Banana or Seedream to generate specific starting images for key video beats, using my Stable Diffusion character shots as reference images.

Then I pick the best of those, tweak them if needed to make them even more visually consistent, and use them as the initial frames of shorter clips that I can later edit together to make a visually consistent longer video.

Lost Cat was the first test drive for this approach, although I really only made a small number of initial images and relied like always on generating a bunch of longer clips and just working with the bits that worked best and that fit together cleanly. I love the result but I spent way more time and money on it than was necessary - the scene I'm re-working right now uses the approach a lot more aggressively. It's more work on the front end but should result in a higher percentage of usable footage and less work on the back end.

The real challenge, I think, is to make the Nano Banana/Seedream clips look and feel as much like my own work in Stable Diffusion as possible, so that the resulting videos to still reflect my personal style.

Files

Comments

Orcus

Very clever. I can't wait to see the results!

Seka

These details help me understand, and appreciate, how much thought and effort you put into your work. Not to mention the technical expertise required to use so many different programs. But all this pays off in the final artwork.

Anaphiel

Thanks - actually putting thought into this is relatively new for me though :) When all I wa doing was making images, then going "hey, I think this would make a great video" and then, well, making it into a video, the hardest I ever had to think was "how do I describe this is a way that the model could interpret"... it's only wanting consistent clips than be edited together into coherent longer videos that has posed a challenge. What I describe here might not be the easiest way to go about that, but it's a way that I think will work and that I personally think I can pull off. We'll see how it goes.

🌸Rose🌸

I absolutely love the vibe 🫠

Orcus

You're being too modest. A hell of a lot of thought and expertise has always gone into your content. As someone who tries to make AI content myself, I could see that instantly. Its probably more now that you're making long format vids, but based on the civitai collection you've clearly been an expert for quite a long time now.