As I begin work on episode 2 of The Ripening, I am beginning to put together concept art for it. Here are some examples of what may be seen in episode 2.

The Ripening is live! A 25-minute sci-fi short film, entirely rendered locally on a single RTX 5060. Set in 2078, decades after a global fertility collapse known as the Silence, the film follows Ember Voss, a young administrative assistant in a surveillance state, as she discovers that her late mother's heritage holds a deep connection to the Tree of Life. Alongside Phoenix, she sets out toward an ancient mystery that may hold the answer to why humanity stopped being able to have children. On a personal note, this has been a long journey. I started tinkering with AI video back in 2024, and the tools have evolved at a breakneck pace since. This is my first real attempt at wrapping a full story into film form. I'm grateful for the many creative friends on X who have inspired me to learn more along the way, and it means a lot to finally share this with you all. Made locally with @ComfyUI,@MiniMax_AI, and @ltx_io

Tomorrow on a screen near you.

Very thorough and well-done analysis by @SamJWasserman. Thanks Sam.

Here is another ComfyUI workflow that I am finding indispensable. It takes two audio inputs. One from the video clip (extracted as audio only), the other from a character reference audio. This ensures consistent voices across multiple video clips. There is nothing worse than having your character sound different across the story.

This is a nice little workflow in ComfyUI for extracting vocal stems from an audio file. Video models often have the annoying tendency of adding music along with the vocals. This separates them.

One man's slop is another man's porridge. There is no good purpose in proclaiming someone's work to be slop. If you don't like it, move on.

Before we start talking about bestowing AI personhood, let's start treating actual living beings as if they matter.

I don't like that Nano Banana now defaults to .jpg instead of .png for image downloads. I do like that they no longer have the diamond Gemini watermark.

I have completed upscaling all 178 clips for RIPENING to 4K resolution. It went quicker than I thought it would. This officially completes "Rough Cut" stage. Following is an outline of the various production stages that I am employing for this film: 1. Assembly - First pass: all clips generated, placed in sequence, basic story order established. Completed. 2. Rough Cut - Reviewing every clip individually for technical/performance issues and redoing/replacing as needed (character fidelity, continuity, aliasing, framing problems, etc.). All identified problem clips redone; all 178 clips upscaled to 4K via Topaz. 3. Fine Cut (current stage) - Full manual rebuild of the timeline in the new 4K project: trimming, pacing, transitions, and any issues that only surface once clips play against each other in sequence. 4. Picture Lock - Timeline frozen; no further cuts, trims, or reordering. Everything downstream assumes the edit won't change again. 5. Audio - Sound mix, sound design/effects, and music placement. 6. Color Correction- Normalize exposure, white balance, and technical consistency across all clips. 7. Color Grading - Creative look pass on top of correction: mood, palette, scene-specific treatment.

These are the Topaz models I am using as defaults for upscaling my film RIPENING to 4K resolution: 1. Proteus - fast precision/GAN model suits the majority of my clips. 2. Starlight Precise 2.6 - slow diffusion model for those clips that need extra quality attention. FYI: Diffusion based models can infer and reconstruct detail using learned understanding of what things (faces, fabric, materials) should look like. GAN-style models enhance detail that's already there. They sharpen, denoise, and clean up existing pixels.

Rough cost comparison between H3/LTX local vs. Seedance cloud video generation for a 20-minute film: Local H3/LTX = $0 (plus negligible electricity overhead) Seedance cloud = ~$3000 Obviously, that is assuming you already have a rig capable of doing the job. If you plan on doing multiple films, it doesn't take long before you are saving 10's of thousands of dollars by going local.

I have everything set with Topaz and ready to roll. Out of 178 clips total, I have 63 for 2x upscale and 115 for 3x upscale. That will hit my target 4K for the entire project. Since I'm doing the upscaling locally, I'll be buried in Topaz for quite a while.