Building a tool for AI series creators - help me get it right!

Hi AI filmmakers :waving_hand:

We’re a two-person team building a production platform for creators making serial work with AI. We really want to understand how you actually work — what you use, what slows you down, what you wish existed.

If you’ve got ~2 minutes, we’d love your take. It’s anonymous, mostly taps - and we’ll share back what we learn.

And a little thank-you: leave a handle or email at the end, and we’ll enter you in a draw for a $50 gift card of your choice - totally optional.

:backhand_index_pointing_right: Building a tool for AI series creators - help me get it right!

Thank you! Every response shapes what we make. :folded_hands:

Hi, I’m Mike DeRosa. I’m trying to create a 45-minute Pilot for a spec 8-episode Limited Series (as a promotional tool to attract investors and producers). I’m using OpenArt.ai to produce 5 to 10-second video clips and then input them into a video editor for camera angles and effects (zoom, pan, etc). My biggest concern is creating consistent characters for my story “Gateway of the Sun” (Three teens find a global network of megalithic teleportation sites, as powerful conflicting forces pursue them). OpenArt’s “Character Lab” sucks. It is NOT consistent. BUT, there is another tool that I’ve just found that lets you train a model and can take up to 500 images (unlike the kiddy Character Lab 2.0, which can only take 4 images). It’s more “pricy” and costs 500 credits per “model,” but I have a yearly subscription, so that’s not a problem for me. I’m just testing out that feature now. Of course it can be pricey if you want to ‘upgrade’ each clip to 4K, (and then there’s Higsfield, which is way more expensive because it generates 4k all the time, but that’s now as ‘character consistent’ or so the reviews go). I plan to generate at 720 and then upgrade the entire video for a one-time fee from an outside source (as yet to be determined). So far, it’s been a fight to get OpenArt to do what I want it to do and not get creative on its own and give me shit I don’t want. I’ll try to keep you updated on what I’m learning. This is new technology, and it’s changing. I started with SORA2, and they shut down the whole thing two months after I got started. SORA2 had a better understanding of spatial geometry. I couldn’t get OpenArt to pan a room the way I wanted. It kept putting windows and doors where they shouldn’t be. SORA2 got what I wanted right off the bat, but it’s gone now. Then I created a charter I liked with the old Character Lab in OpenArt with 20 images, and then it pulled the rug out from under me and changed the interface to the new Character Lab 2.0. and only used the first 4 images I had uploaded, and my new generations were NOT consistent (with uniforms and clothing that didn’t match). So when I say the technology is changing fast… It’s changing AS you use it! Keep in touch. There’s strength in unity. YES, enter me in that contest. MY email is: mdsungate@gmail.com Put something in the subject line like “Anna from AI series creators”, so I know it’s not more junk mail. I wrote a novel about the same story, “Gateway of the Sun: A Novel,” and I get so many stupid emails from scammers. Best Regards, Mike DeRosa

Mike, I hit this exact wall making a dark fantasy anime on my own. What actually helped wasn’t finding a better consistency tool, it was lowering what I asked for. I stopped trying to lock the character 100% and locked 3 or 4 things instead: silhouette, color palette, one detail I never let change (a scar, a specific collar, the hair shape). If those hold, people read it as the same character, especially once the cut and the sound are doing their job. Chasing a perfect match frame to frame usually wrecks the pacing anyway.

Two things that helped in practice:

  • Pick a few reference stills you’re actually happy with and reuse those as the source for every shot of that character, instead of generating from text each time. You get consistency by limiting the model, not by re-rolling until it behaves.
    • For your room problem (windows and doors moving around), build the space once as one reference image and treat every new shot as a reframe of that same room. When you let it start from a blank prompt it just reinvents the geometry.
  • On tools changing under you: I keep the character sheet and the approved frames in my own folders, not inside one platform. So when an interface changes or a tool shuts down (Sora did that to me too), the character survives the move. Happy to swap notes on the series side, keeping it consistent across episodes is the part nobody warns you about.