← sources

H3 Test Render - 35 second tracking/walking shot.

source: reddit (reddit:StableDiffusion)published 2026-10-18 17:19ingested 2026-10-19 02:40

retained text

Aspiring filmmaker here, probably older than most of you haha.

I started off experimenting with "AI art" with Disco Diffusion a number of years ago doing some animation stuff. Took a break for a while from AI largely due to the anti-AI backlash from a lot of my creative peers.

The advances in video/animation pulled me back in because I feel that it's at a point where I can start creating the movies that I always wanted to make. I've always had the skill to direct, but I lacked the skills in networking and fundraising that you *also need* to make traditional films.

I've had pretty good luck with render quality and consistency, but the one thing I really wanted was *longer shots.* 15 seconds or so is generally ok for a lot of scenes as most individual takes in a cut scene don't last much more than 5 or 10 seconds. *But...* Sometimes you want/need something a little longer. Like think of the effectiveness of the tracking shot in *Goodfellas*.

[https://youtu.be/kpvNjbtF9Hg?si=wagSTLVN3NbaKswp](https://youtu.be/kpvNjbtF9Hg?si=wagSTLVN3NbaKswp)

I decided to do a "SciFi" test of a technician walking through a narrow corridor of electronic equipment, pausing to press a button and then continuing to walk.

The shot is 35 seconds long. It was rendered out at 1.6 MP, 20 steps using Singularity as the model. It used a 5 panel image of her face and a separate 5 panel image of her body as reference images for her look... left and right profile, left and right 3/4 view, front view.

I rendered through [RunPod\* ](https://runpod.io?ref=gmm1tk3l)on an RTX Pro 6000 with 96 GB of VRAM. I used a custom patched version of Sage Attention because at a certain size, Sage causes the render to look like what I call "brown snow." This "pod" has a current cost of $2.49/hour plus a minimal cost for storage container which is only cents per hour. My set up's container cost is $0.035/hour.

I then took the render into DaVinci Resolve and used its "AI Superscale" set at 4X RTX nVidia. In the color module, I did a slight color correction to adjust the contrast, noise reduction, and then added a film grain at the end.

The video above is the result of all that. :-)

\* The RunPod link is a referral link. I first used them during my Disco Diffusion experiments in 2022 when they were just getting started as a business and found them really awesome.

If you use the referral link to sign up, you'll get a $5-500 credit. Anyway... I'm blabbing too much about this.

[https://runpod.io?ref=gmm1tk3l](https://runpod.io?ref=gmm1tk3l)

***<Edit>*** Snipped my long story *(thanks ADHD!)* about using RunPod as it was reading too much like a full on ad so I'm removing it. I'm leaving the referral link because some might find it useful. Full disclosure, this is what I get out of the referral if someone signs up using it: "A one-time credit from $5-500 when a user signs up with your link and adds $10 for the first time. Credits on referred user spend during their first 6 months. (5% Serverless and 3% Pods)"

***Thanks to everyone here for their tips and advice which so helped me get back up to speed with current creative "AI generation." I really appreciate it!***

***May you all have wonderful and error-free renders!***

document fetched 2026-10-19 00:29 · canonical https://reddit.com/r/StableDiffusion/comments/1x98r2r/h3_test_render_35_second_trackingwalking_shot