"Dark duels in the Twisting Grove" update: story board animations WIP

Here's an update on my experiments with GAI for comic animation—what worked, what didn’t, and how I adjusted my approach. TLDR: Use Wan2.2Fast on the cheapest settings to test ideas and not waste creds. It gives you the broad motion and timing cues you need to refine prompts; save Pixverse V6 for the final polish. This is a quick, candid update on my experiment using GAI to make animations for a comic page. I wanted to see whether the cheapest, fastest model could be a useful iteration tool before burning credits on higher-end renders. I ran a handful of tests with one storyboard reference and an animator-style prompt that tried to animate the whole page as a single wide-shot, frame-by-frame — ambitious, messy, and exactly the kind of thing I should have expected to trip up the cheap model. What I did One storyboard reference image; one animator-style prompt aimed at animating the entire page as a continuous wide shot. Animations 1–4: Pixverse V6, 5s clips at 360p. Animations 5–7: Wan2.2Fast, cheapest settings. Goal: validate concept and prompt language cheaply, then refine before committing credits. What I saw (practical takeaways) Wan2.2Fast on cheap settings actually gave me a useful early signal. It reproduced the direction of movement and rough timing well enough to tell me whether the idea was worth pursuing. That’s the key: it’s not about final quality, it’s about whether the motion intent reads. Pixverse V6 looked cleaner and more coherent, but the difference in movement intent wasn’t always worth the extra credits during early exploration. Prompt wording mattered a lot. Small tweaks exposed weak or conflicting directives that pulled the animation off course. Overly prescriptive or compound instructions tended to confuse the cheap model; concise, motion-focused language worked better. Iterating on the prompt helped me isolate which phrases produced stable motion and which introduced distracting artifacts. Where I screwed up (self-critique) I was too ambitious with a single wide-shot approach. The storyboard complexity amplified failure modes in the cheap model. I should have started with smaller, modular panel tests. Sample size was tiny and subjective: I eyeballed results instead of running multiple seeds or trials to measure consistency. I didn’t control for randomness or seed variability, so some “failures” might just be bad draws rather than model limitations. Practical recommendations Use cheap models for rapid prototyping of timing and blocking. Break the page into smaller shots or single-panel animations for clearer diagnostics. Iterate prompts aggressively; remove conflicting directives. Run multiple seeds before deciding a model is unusable. Save high-cost renders for final polish. This is a pragmatic, slightly embarrassed note for other GAI animators: cheap models are great for quick checks, but design your tests to isolate motion intent from compositional complexity.