19:43
AI video generation
Olivio Sarikas presents a set of reusable ChatGPT Image 2 + Seedance 2.0 (via Runway) prompting workflows — including a "juxtaposition" story-generation trick, motion/story/storyboard grids, and character/world reference sheets — that let a creator build a fake video-game trailer, a choreographed two-character fight animation, and short cinematic films starting from as little as a single source image.
A fake video game trailer can be built from a ChatGPT Image 2 start screen, settings screen, and in-game scene, then animated in Seedance 2.0 without time codes so Seedance decides its own pacing.
A 4x4 movement grid (16 poses with motion arrows), adapted from a prompt originally created by "Ugi" on X for dance motion, lets Seedance 2.0 animate a scripted fight sequence between two characters in the exact 1-16 order.
The "juxtaposition" technique places two opposite character archetypes (punk/sophisticated lady, monster/nun, vampire/priest, etc.) together so ChatGPT Image 2 invents a title, description, and narrative connection between them.
A juxtaposition can be expanded into a 6x6 grid of 36 narrative beats showing how a story's arc rises, though the video states these steps are too random/numerous to film directly and function as a planning layer.
A 12-scene (two rows of six) or 3x3 nine-scene storyboard grid converts a juxtaposition or story arc into consistent, visually distinct, filmable scenes with a real story arc.
Character item sheets (front/back plus individual props like a pocket watch, gloves, top hat, ruby ring, leather shoes) on a white background give the AI the detail needed for consistent close-ups.
A character bio sheet (age, job, city, height, weight, eye color, hobbies, strengths, weaknesses, backstory) can be auto-generated by ChatGPT Image 2 from a minimal prompt and reused to keep a character consistent across a project.
A world-building sheet combines a landscape/civilization panorama, three characters, weapons/vehicles, and a map, with elements swappable (tools instead of weapons, ships instead of vehicles) to fit the setting.
A single existing image (e.g. a Midjourney soldier-and-girl image) can be expanded into a full short film by extracting character sheets, building a 3x3 storyboard, and generating a time-coded Seedance shot script from the storyboard image.
Runway's Seedance 2.0 settings let users pick resolution and length second-by-second (previously fixed at 5/10/15), toggle sound, and generate for free in an unlimited/Explore mode instead of spending credits.
Two bonus non-film tricks: turning a plain smartphone food photo into a professional product design via a minimalist prompt, and generating "exploded assembly drawing" style technical illustrations.
Fake Video Game Trailer Workflow — A four-stage workflow — a ChatGPT Image 2 start screen, settings screen, and in-game scene, plus a Seedance 2.0 video prompt — used to fabricate a realistic-looking video game trailer from still images. Apply: Generate a start screen (menu buttons like start/settings/credits/multiplayer), a settings screen (character on a throne-like chair choosing powers), and an in-game scene (with UI) in ChatGPT Image 2, then feed the images into Runway's Seedance 2.0 with a motion prompt that omits time codes so Seedance sets its own pacing.
Movement/Motion Grid — A 4x4 grid image (16 poses) with directional arrows showing a character's sequential movements, originally devised by "Ugi" on X for dance motion and adapted here for a two-character fight. Apply: Generate a 4x4 grid of 16 numbered poses for two characters (e.g. ninja vs. Viking) with motion arrows in ChatGPT Image 2, then instruct Seedance 2.0 to animate strictly following the 1-16 sequence with dynamic camera angles.
Juxtaposition Technique — A creativity method where two opposite/incompatible character archetypes (e.g. a punk and a sophisticated old woman, a monster and a nun) are placed side by side so ChatGPT Image 2 invents a title, description, and narrative connection between them. Apply: Prompt for a highly detailed 3D character image showing two opposite characters on a white background with a title and description explaining the juxtaposition, then use that pairing as the seed for a story arc or storyboard.
Story Arc Grid (36-step) — A 6x6, 16:9 grid image that expands a juxtaposition into 36 narrative beats (e.g. "chaos incarnate," "the shattering discovery," "the first collision") showing how a story's arc rises over time. Apply: Give ChatGPT Image 2 a short summary of the two juxtaposed characters' motives and ask for a 6x6 grid of 36 steps; use it as a pacing/planning tool rather than literal shots, since it has too many/too random scenes to film directly.
Storyboard Grid (12-scene / 9-scene) — A two-row-by-six (12 total) or 3x3 (9-scene), 16:9 grid of consistent, visually distinct, expressive scenes forming a usable storyboard for a short film. Apply: Prompt ChatGPT Image 2 for a storyboard of a set number of scenes with visual difference/expressiveness per scene and a coherent story arc, then convert it into a Seedance 2.0 prompt requesting dynamic, cinematically connected camera angles.
Character Item Sheet — A character reference sheet showing a character's front and back plus individual prop/clothing items (e.g. pocket watch, gloves, top hat, ruby ring, leather shoes) on a white background. Apply: Generate the sheet with a "highly detailed 3D gaming character" prompt listing the outfit, front/back views, and individual items, then reference specific items (e.g. "close-up of the hand with the ruby ring") in later prompts for consistent renders.
Character Bio Sheet — An auto-generated character profile (age, job, city, height, weight, eye/hair color, hobbies, strengths, weaknesses, backstory) produced alongside the character image by ChatGPT Image 2 from a minimal prompt. Apply: Prompt for a character image plus a listed set of bio fields (name, age, job, city, size, weight, description, hobbies, strengths, weaknesses, backstory) and let ChatGPT Image 2 fill or extend them, then reference the bio in later prompts for consistent characterization.
World-Building Sheet — A combined single-image layout containing a landscape/civilization overview, three characters from that world, weapons/vehicles, and a map (e.g. for city planning). Apply: Prompt ChatGPT Image 2 for the combination and swap elements to fit the setting — e.g. tools instead of weapons, ships instead of vehicles, or move the panorama to a side layout for an interior/city-only scene.
Single-Image-to-Short-Film Workflow — A pipeline that starts from one existing image (e.g. a Midjourney image of a soldier and a girl) and expands it into character sheets, a storyboard, and a full Seedance video via ChatGPT Image 2 and Runway. Apply: Take a minimalist-prompted source image, use ChatGPT Image 2 to generate detailed front/back versions of each character plus supporting characters, build a 3x3 storyboard grid with dynamic connected camera angles, then generate a timestamped Seedance script from the storyboard image.
Time-Coded Seedance Script Generation — A method of feeding a generated storyboard image back into ChatGPT and asking it to write a full shot script (time code, camera angle, action, movement) for a 15-second Seedance video. Apply: Upload the storyboard image to ChatGPT and prompt for a 15-second script with time codes covering camera angle, action, and movement, then copy the resulting script into Seedance 2.0.
White-Background Isolation Technique — A prompting convention of generating character/item sheets on a white background to simplify separating the subject from its background and prevent the AI from confusing background color/brightness with character elements. Apply: Append "on a white background" to character and item-sheet prompts whenever consistency or later background separation is needed.
Explore/Unlimited Mode (credit-free Seedance generation) — A Runway generation mode that, per the video, works with Seedance 2.0 at 15 seconds/1080p and lets users generate video without spending credits. Apply: Switch Runway to "unlimited"/"Explore" mode when generating Seedance 2.0 clips to produce iterations for free instead of using paid credits.
Smartphone-Photo-to-Professional-Design Prompt — A minimalist-prompt technique that transforms an ordinary smartphone photo (e.g. of food) into a professionally composed design with icons, custom fonts, and text layout. Apply: Feed a plain smartphone photo into the image model with a short prompt asking it to turn the photo into a professional design, letting the AI decide composition, color, and typography.
Exploded Assembly Drawing — A named illustration style — a "3D highly detailed exploded assembly drawing" — that depicts a technical object (e.g. a headset) with its components separated apart. Apply: Use a minimal prompt such as "a cool-looking design of a 3D highly detailed exploded assembly drawing of a [object]" to generate this style.
The video frames using deliberately minimalist prompts and letting ChatGPT Image 2 infer details (character bios, 36-step story arcs, juxtaposition descriptions) as a time-saving strategy that also increases consistency, not as a shortcut around quality.
The juxtaposition technique is presented as a general creativity method — placing two incompatible archetypes in proximity forces the model to invent a connecting narrative — independent of any single project.
Omitting time codes from the Seedance prompt is a deliberate choice to let Seedance figure out its own pacing; time codes are reserved for when the creator wants tighter control, as demonstrated later by generating an explicit timestamped script from a storyboard image.
White backgrounds are used specifically because they simplify background separation and prevent the AI from confusing background color/brightness with parts of the character — an explicit segmentation technique, not a stylistic default.
Generic prompts for well-known archetypes (e.g. "Viking") are called out as a stated limitation: ChatGPT Image 2 defaults to "the most popular Viking on the internet" unless the prompt adds distinguishing detail.
Explicitly requesting "very dynamic camera angles" is necessary to avoid a static, side-on "Street Fighter" framing in fight/action storyboards — a concrete negative case the video calls out.
The 36-step story arc grid is explicitly distinguished from a shootable storyboard: too many/too random scenes to film directly, but valuable for seeing how the narrative arc and pacing rise over time.
For the single-image project, rather than hand-writing the Seedance shot list, the creator fed the finished storyboard image back into ChatGPT and had it generate the full timestamped script (camera angle, action, movement) automatically.
«How to make a fake video game precise fight animations making a storyboard out of just one image?»
— 00:00
«C Dance two 15 seconds 1080p also works in the unlimited or explore mode where you don't use any credits.»
— 03:19
«this is something that really a little bit annoys me about Chat GPT image two that when you have a term like Viking it's creating the most popular Viking on the internet.»
— 03:54
«Now I have to be honest this very long prompt, very detailed here is not from me. The original prompt actually comes from Ugi over on X.»
— 04:46
«if you don't have that in your prompt what will happen to you is that you have a Street Fighter kind of situation where the characters are just filmed from the side.»
— 05:41
«So what I do here is check the position. So you come up with two opposites that don't really mix together but you place them in proximity and a story is created through that.»
— 06:21
«these are not really scenes you can actually use to create your film because they are kind of too random but when you look at these 36 steps of a story they actually explain how the arc is rising, what is happening over time.»
— 07:30
«Again, the white background is something I use because then I have a easy background separation and there is nothing that interferes from the color or the brightness where the AI might confuse something in the background with something that is part of the characters.»
— 12:05
Reception
Overwhelmingly positive reception with viewers praising the guide's quality and helpfulness; most engagement comes from gratitude and learning, with minimal substantive criticism.
The video functions as a dense prompt-recipe library rather than a deep explainer, cycling through eleven-plus discrete named techniques (motion grids, juxtaposition, story-arc grids, storyboards, character and world sheets) with working prompts but only brief commentary on why each works, leaning on the shared slideshow to carry replication detail.
19:43