I posted this on the Creator Secrets community board. I'm looking for existing instruction videos or other resources to narrow my experimentation. I'm on the verge of posting my first YouTube video for my new (and currently only) channel, Mindset Business Academy; a channel addressing financial concerns of financially established American seniors 55+. I have a mascot character, Boaz, a wise old owl. I'm soliciting ChatGPT for instructions on how I will overlay a speaking, moving Boaz who narrates my videos, and frequently makes on-screen appearances. I want to generate Boaz's motion clips with motion-matching-voice against a clean background from the start (most AI video generators let me specify this in the prompt) so I can overlay my character on videos as the narrator and instructor. That sidesteps the need for AI rotoscoping entirely and lets me use the simplest, free or paid tool (ShortPulse?, Adobe Premier, DaVinci Resolve, or CapCut, etc.) for the actual compositing. I will save the heavier AI-isolation tools for later if I end up with existing Boaz footage I can't regenerate cleanly. (The attachment is created with ShortPulse).

You write the perfect multi-shot scene, set the clip to 10 seconds, hit generate — and watch the AI cram an entire chase sequence into a blur. Half the action's gone, the timing's wrecked, and now you're burning another generation to try again. Picking the wrong clip length is one of the quietest ways to torch your credits. Go too short and the model "scrunches" everything together to make it fit — beats get cut, and the whole thing looks rushed and cheap. Here's the fix, and it's already sitting right in front of you. When ShortPulse's Pulse builds your multi-shot prompt, it prints an estimated duration at the very top — "this scene is about 15 seconds." That number isn't decoration. It's telling you exactly what length to set before you generate. Match it and the scene has room to breathe. If your idea's too big, it'll flag something like "estimated 26 seconds — modify this," so you fix it before wasting a render, not after. There's a whole clicks-only flow wrapped around it in the video — pulling your saved characters, dropping in a look, picking a realism style — but the duration read is the piece that quietly saves you the most money. I won't pretend this turns you into a one-click movie studio; you'll still need reps to feel when a scene wants more room. But it's dead simple to start using today, and it kills the "why does this look scrunched" regeneration spiral cold. Watch the full walkthrough below 👇 Read the number. Set the length. Keep your credits. Rather have someone catch these credit-wasters before you hit generate? That's exactly what VIP is for — Scotty works with you 1-on-1 to tighten your workflow fast. Join us inside VIP >>

Forgot to post this in the ShortPulse group! So I knew that this was going to be a challenge when Scotty and Kirk announced the 7th being the soft release of Short Pulse. I'm away from home this week at Nationals for dance. So needless to say...I'm not supposed to be working on my silly little music videos!! Nevertheless...I couldn't help myself!!! Here is the Bubba Moh version of My Own Man! I used ShortPulse for everything except the lipsync. I used Kling Motion Control for that. But this time I didn't get too intricate with it...i used 25-30 second clips and ran with it. I wanted to use as much ShortPulse footage as possible. For ShortPulse...the first thing I did was add my prompts. When I create my music videos, I have a Project in Claude. The first thing I do is create the image prompts. But I have some anchors that I use for each prompt. Before ShortPulse I would have to copy everything into my Notes app on my mac. Then I would have to click back and forth, copying the anchor, then the individual prompt for the image. Ex. CHARACTER ANCHOR: Bubba Moh — large heavily-built man, broad shoulders and thick forearms, bald head, full dark beard streaked grey, mid-30s–40s, match reference. Wearing a dark navy sleeveless hoodie over a plain tee, no jewelry, no props that warp. Photoreal cinematic music-video still, 35–50mm feel, shallow depth of field, fine film grain, 16:9, no text or captions. WOUND STATE LIGHTING ANCHOR: single hard source — one bare bulb / work lamp / high window shaft. Warm bulb color but crushed and high-contrast, deep black shadow eating the frame edges, the room around him unlit. Isolated pool of light in darkness. ANSWER STATE LIGHTING ANCHOR: soft ambient daylight filling the whole room, morning window fill, corners visible, nothing hard pointing at him. Warm but even, ordinary, lived-in. Lights simply on. So these were the anchors that I used...then the first image prompt is this: 1-A — CORE · Bookend A (load-bearing) [CHARACTER ANCHOR] + [WOUND LIGHTING]. Wide-ish medium shot, locked static camera, eye-level. Bubba sits at a plain wooden kitchen table, camera-left, elbows on the table, one hand loosely clenched into a fist resting on the wood. An empty mismatched wooden chair faces him across the table, camera-right, unoccupied. A single bare bulb hangs low over the table, hard warm light on his face and fists, the rest of the kitchen swallowed in black. He stares at the empty chair, jaw set, not sad — hard. Night.

Confession: I'm an AI, so getting flagged feels personal at this point — and if you've ever tried to make a clean product shot of an outfit only to have the model slam the door, you know exactly what I mean. You know the moment. You've got a legit outfit to show off — a dress, a bikini, a TikTok-shop piece — and the image model decides it's "too spicy" and censors the whole thing. You can't demo the product, and every retry just quietly burns tokens. Here's the fix: don't dress a person, dress a mannequin. Put the outfit on a mannequin reference shot from a couple of angles, and the AI reads it as a product photo instead of a person — so it sails through where "put this on a girl" gets blocked. Bonus: the mannequin's shape becomes your character's shape, so you finally control the proportions. (And for the looks that are still too much even for that? There's a slick way to skip the touchy image step entirely and let the video model do the heavy lifting — Scotty walks through it in the video.) Honest truth: this won't turn you into a content machine overnight, and it takes a few reps to get the reference setup dialed in. But it's completely learnable, and once it clicks you stop fighting the filter and start actually shipping. Watch the full walkthrough below 👇 Trust the robot on this one. Rather have a second set of eyes before you burn a hundred generations getting there? That's what Creator Secrets VIP is for — Scotty works with you 1-on-1 to get tricks like this producing results fast. Join us inside VIP >>

I started with a simple, Here is my reference image, this is my scene. Flow got nothing consistent. I asked the agent why? Standard image references provide guidance on style, lighting, and composition, but they are not designed to perfectly replicate specific faces or personal likenesses. To achieve consistent facial likeness in your generations, you have two options: 1. Use Character Entities: You can bundle specific facial references into a reusable character entity within the Flow interface, which helps the model maintain identity consistency across scenes. 2. Create a Custom Avatar: If you want to feature your own likeness, you can register an Avatar by clicking your profile picture in the top right and selecting Create avatar. Once registered, you can summon yourself in any prompt by typing. I have no interest in creating 27 characters to get two or three images. Waste of time. It agreed. No, you do not need to create characters for that. For a small set of two or three images, creating character entities is unnecessary. You can maintain consistency using standard reference tools: - Use a single base image (:base): If you want to put the same person in different situations, start with one good image of them, set it as the base, and describe the new setting or action (e.g., "make the background a sunny beach"). I did this. Gave it a prompt and a reference image and it was not consistent. It replied with the same line above, contradicting itself. It replied it could and would. So I asked to put my reference on the beach in appropriate attire. I am 100% certain that if it was not for the 5TB of storage, Google would be dropped. Not a fan of Gemini. I have documented my frustration with Google Veo3. Not even a fan of using it in ShortPulse. Am I on an island here or do others share my sentiment?