5 YouTube videos that take you from your first AI image to your first AI film
Hey, it's Cindy 馃尡 Here are the five videos from the reel, in order, with the timestamps worth jumping to.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
What is free and what is not. All five videos are free to watch. The tools are not all free to use. Video one runs on ChatGPT, and image generation is available across ChatGPT plans, so that one costs nothing. Videos two to five run on Higgsfield, which is a paid platform with credit-based plans. Check the current plans on their site before you commit. Watching video two first is the sensible order precisely because it shows you what you would be paying for.
why this order
Still image, platform, you, motion, story
Each one only makes sense once the last one has landed. Prompting a still image teaches you the language. The platform tour teaches you the map. A consistent version of yourself gives you a character. Motion gives the character life. The short film puts all four together.
Total is 1 hour 43 minutes Five nights, not one sitting. Generating something after each video is what makes the next one stick.
Stop after video three if you want Videos one to three get you photos of yourself that look like you. That is a complete outcome on its own.
The prompting fundamentals, with his actual prompts on screen rather than described. The most useful part is the editing conversation: uploading a real photo and changing it by talking, rather than writing one perfect prompt and hoping. That habit transfers to every tool in the rest of this list.
You walk out with: the structure of a prompt that produces what you pictured, and the ability to fix an image by talking to it.
One note. It was recorded before the most recent ChatGPT image updates, so the interface on screen looks a little older than yours. The prompting itself has not changed. Image generation and editing an uploaded photo are documented across ChatGPT plans in OpenAI's own help page.
The full platform in one pass. Watch this before you pay for anything, because eighteen minutes tells you whether the feature set matches what you actually want to make. It covers the image side, the video side and how the pieces connect, including running it from a chat rather than the web interface.
You walk out with: a decision about whether Higgsfield is worth it for your use case, made on evidence rather than a thumbnail.
The shortest video on the list and the one with the biggest payoff. It builds a Soul ID character, which is a trained likeness of you that stays consistent across every image and video you generate afterwards. This is the step that separates "a nice AI picture" from "a photo of me that I can actually use".
You walk out with: a character of yourself you can shoot with forever, rather than a one-off image.
Get the training photos right first. This is the step everything downstream depends on, and it is almost entirely about the photos you feed it. There is a checklist below.
This is a craft video rather than a tool video, which is why it is the one worth watching twice. It is about why AI video reads as AI, and the specific things that fix it: imperfection, real dialogue, camera movement, sound design and editing. Almost none of that is about which model you use.
2:12 路 What makes a video realistic. The framing for everything after it.
3:18 路 How to create imperfection. The single highest-value idea here. Too clean is the tell.
7:22 路 The importance of dialogue. Why silent clips feel generated and spoken ones do not.
13:18 路 Camera movement. Motion that implies a camera operator, rather than a floating lens.
16:10 路 Sound design, then 18:48 editing. The last twenty percent that most people skip entirely.
You walk out with: a checklist for why your clip looks fake, and the specific fixes.
The longest one, and the only one that shows a complete production rather than a technique. Script, assets, four scenes, then the finished film. Watching someone go from a blank page to a film in one sitting is the thing that makes the whole workflow click, and the scene chapters mean you can skip to whichever part you are stuck on.
4:31 路 The script stage. Written with Claude. If you only watch five minutes of this video, watch these.
5:25 路 Assets. Building the character and the world before generating a single shot.
12:42, 19:53, 24:37, 31:31 路 The four scenes. Each one is a different problem. Watch the one closest to what you are making.
35:03 路 The finished film. Worth watching properly, because it sets the standard the rest of the workflow is aiming at.
You walk out with: the whole production shape, from idea to finished cut.
A note on the title. YouTube's own title for this one uses a long dash in the middle. It is the same video.
do this first
The training photo checklist
Video three is the video everything else depends on, and it lives or dies on the photos you upload. Higgsfield's own help centre says a Soul ID character needs 20 or more photos of the same person, up to 80 supported. What it also says, and what people skip, is what those photos have to look like (source).
馃摲 20 photos minimum. More is better up to 80. Twelve good photos produces a worse character than thirty average ones.
馃攩 Well lit, different angles, different expressions. Front on, three quarters, profile, smiling, neutral. A set of twenty near-identical selfies teaches it one pose.
馃毇 No sunglasses, no heavy shadows, no cropped faces. Anything that hides part of your face teaches it to guess that part.
馃 At least one full-height photo. Required, and the reason generated full-body shots either work or come out strange.
馃檯 Only your own face. Or someone who has actually agreed to it. That is their rule and it is a good one.
Spend the twenty minutes. Gathering the right thirty photos is genuinely the difference between a character that looks like you and one that looks like your cousin. Everything after this step inherits whatever you feed it here.
night one
Make something tonight
Before you watch anything, make one image, so the tutorials land on something you have already tried. Open ChatGPT, attach a photo of yourself, and paste this.
馃帹 my first proper image prompt
Using the photo I have attached as the reference for my face, generate one photorealistic portrait of me, 4:5, in a warm-lit indoor setting with a large window just out of frame to the left.
Keep my face, features, skin tone, skin texture and proportions exactly as they are in the photo. Do not slim, reshape, lighten or smooth anything. Real skin texture, not retouched.
Lighting: soft directional daylight from the left, gentle fill on the right so the shadows stay open, warm neutral background with enough depth that I am clearly separated from it. Shot on an 85mm lens at eye level, shallow depth of field, focus on the eyes.
Then tell me, in three short lines, which parts of this prompt did the most work, so I know what to change first next time.
The last paragraph is the point. Asking it which parts of the prompt mattered turns one generation into a lesson. Do that on every early attempt and you will stop guessing within a week.
five nights
A schedule that finishes
Night one, 14 minutes. Video one, then run the prompt above and change one thing about it three times.
Night two, 18 minutes. Video two. Decide whether you are paying for the platform before you go further.
Night three, 8 minutes plus prep. Gather your training photos against the checklist, then video three, then train the character.
Night four, 23 minutes. Video four, then generate one short clip and deliberately break the "too clean" rule to see the difference.
Night five, 40 minutes. Video five. Then write one scene of your own, even a bad one, while it is fresh.
You can stop at night three. Videos one to three give you a complete outcome on their own: consistent photos of yourself that actually look like you. Nights four and five are for when you want them to move.
the waitlist is now open
AI System Club is launching soon
Build AI systems with me through recorded lessons you can follow at your own pace. Join the waitlist for first access and the founding price.
btw this list gets first dibs the moment doors open :)