I Made a Cinematic Ad Using Claude Fable 5 + Higgsfield AI (Full Workflow)

2026-06-15 · case-study tutorial · Adil · recording, 46:05 · watch on YouTube

Topic: Video generation & shot direction (also: Assets, references & consistency, Prompting & AI assistants, Audio, voice & music) · mode: Tutorial · tags: prompts shared, fiction, one creator's experience

Adil makes a cinematic football ad for the TOPUP soda can in three stages. He builds and tests the assets first, with Soul Cinema for his likeness and the location, then adds each as an Element named exactly as in his Claude Skill, which writes the prompts with a shared style header. Across twelve scenes he fixes misses with shot-by-shot notes to Claude, and pins the wandering goals with a drawn location scheme: "a map beats a paragraph". His recap: the final film is the best few seconds of many tries, so iteration is the skill.

What is said

The parts of the video and what is said in each (★ — how much it matters to the video).

00:00:00 — Intro: from halftime commercials to one website; the three-step workflow

★★ Adil shares the system prompt (a Claude Skill) that the Higgsfield team spent months perfecting and that does most of the heavy lifting of prompting the TOPUP football ad. — 00:00:24 · Claude · also in: Make a Cinematic Ad End-to-End: Intro: One Website, One Cinematic Ad
★★ Adil's ad workflow has three steps; step one is the assets: characters, locations and props made in Soul Cinema and GPT Image 2. — 00:00:24 · Soul Cinema, GPT Image 2

00:00:59 — The finished TOPUP commercial played in full

00:02:45 — Stage one: the idea, the can sheet, Adil's character sheet, the New York street, the asset test, two robots, the ball and the goal

★★ Adil works in a separate Collab project, which gives access to all image and video models in one place and keeps everything made for the video together and easy to share with the team. — 00:03:07 · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★ Adil calls GPT Image 2 the best model right now for image editing. — 00:03:07 · GPT Image 2
★★ Adil makes a product sheet of the TOPUP can in GPT Image 2 so that the video model knows the can from all angles and does not hallucinate it. — 00:03:51 · GPT Image 2
★★★ For a character with his own likeness Adil calls Soul Cinema hands down the best model, keeping all facial features consistent: he selects Character and drops in 20 images of himself. — 00:03:51 · Soul Cinema · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★ In the character-sheet prompt Adil describes only the layout (two panels: full body on the left, face close-up on the right) and leaves the outfit open so Soul Cinema offers outfit options. — 00:04:19 · character sheet, Soul Cinema
★★ Adil says eight Soul Cinema images cost one credit, so five credits buy many outfit options, which makes the character sheet a step to experiment freely. — 00:04:47 · Soul Cinema, character sheet · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★ Adil picks a casual outfit for his character to contrast with the massive futuristic robot made later. — 00:05:16 · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★★ Adil calls the location the most important image: a sloppy or plasticky one ruins the videos whatever the prompt, since the video takes textures and lighting from it; Soul Cinema gives movie-still looks by default. — 00:05:16 · Soul Cinema, AI look · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★ Adil sets the ad on a New York street, saying he was there just the week before. — 00:05:16 · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★ In the Soul Cinema location prompt ("cinematic anamorphic lens, shallow depth of field, film grain"), the anamorphic lens gives the wide cinematic feel and film grain removes the too-clean digital look. — 00:05:58 · Soul Cinema
★★★ Before locking key assets Adil tests them together, here the character on the chosen location, and calls this stage critical. — 00:06:25
★★ For the asset test clip in Seedance 2.0 Adil generates only five seconds and a single generation. — 00:07:22 · Seedance 2.0 · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★ Adil has Claude write the asset prompts because it writes detailed, structured prompts for him. — 00:07:22 · Claude · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★ Adil asks Claude for a robot character-sheet prompt (two full-body views, front and back, white and red armour, number seven on the chest) and runs it in Soul Cinema. — 00:07:58 · Claude, character sheet, Soul Cinema
★★ With the prompt enhancer on, Soul Cinema gets very creative and returns a wide variety of robots. — 00:08:23 · Soul Cinema
★★ Adil makes the rival robot in the same style as the hero's robot so they look like they belong in the same world, changing only the number (nine) and the colours. — 00:08:23
★★ Adil says props matter as much as characters in long videos, since they too have to stay consistent across all scenes and shots. — 00:09:39 · also in: Make a Cinematic Ad End-to-End: Stage 1: Building the Assets
★★ Adil says prompting is where most people fail: they overload their prompts with unnecessary information and waste time and credits. — 00:10:48 · also in: Make a Cinematic Ad End-to-End: Stage 2: The Prompting Framework

00:11:14 — Stage two: installing the Claude Skill and giving Claude the script, the assets and their names

★★ The Claude Skill packs Higgsfield's most recent learnings on crafting prompts, so users skip the painful loop of editing and retrying prompts. — 00:11:26 · Claude · also in: Make a Cinematic Ad End-to-End: Stage 2: The Prompting Framework
★★ To install the downloaded skill in Claude: Customize, then Skills, the plus button, then Upload a Skill. — 00:11:51 · Claude · also in: Make a Cinematic Ad End-to-End: Stage 2: The Prompting Framework
★★ In a new Claude chat Adil attaches the script as a PDF and all the assets, and types each element's name with a short description to give Claude full context. — 00:11:51 · Claude · also in: Make a Cinematic Ad End-to-End: Stage 2: The Prompting Framework
★★★ In Higgsfield Adil adds each asset as an Element and names it exactly as in Claude, which he calls the key step. — 00:12:57 · Elements, Claude · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation

00:13:09 — Scenes one to five: Elements, the floating can, the transformation, the robots meet, the goal is born, the first duel

★★★ Every prompt of the skill carries the same style header, locking lighting, lens, acting style and physics across all shots. — 00:14:29 · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation
★★★ Adil's insight for video prompts: ask for no music and only environmental sound effects, because the music is added over the final video; the skill already includes this. — 00:14:29 · native audio · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation
★★ Pasting a Claude prompt that names the Elements into Seedance 2.0 attaches them automatically, with nothing attached by hand. — 00:14:55 · Elements, Seedance 2.0, Claude · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation
★★ Adil's Seedance 2.0 settings for the ad scenes: 16:9, 1080p, 15 seconds. — 00:14:55 · Seedance 2.0
★★ Seedance 2.0's first takes of the transformation scene broke the hero's smile and had him running backwards. — 00:16:46 · Seedance 2.0 · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation
★★ To give the transformation impact, Adil asks Claude for super macro close-ups of the robot's armour pieces locking on. — 00:18:17 · Claude · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation
★★★ The rhythm of Adil's workflow: problems are not fixed by hand; you watch the takes, give notes and Claude rewrites the prompt. — 00:18:46 · Claude · also in: Make a Cinematic Ad End-to-End: Scenes 1-2: The Opening & Transformation
★★★ When a generation misses, Adil stops giving Claude vague notes and directs shot by shot, describing what the camera sees in each shot (close-up, over-the-shoulder, low handheld angles) rather than what happens. — 00:20:36 · Claude · also in: Make a Cinematic Ad End-to-End: Scenes 3-4: First Meeting & The Goal
★★ From a plain description of the robots' first meeting, Seedance 2.0 gave a static wide shot with the robots barely moving. — 00:20:36 · Seedance 2.0 · also in: Make a Cinematic Ad End-to-End: Scenes 3-4: First Meeting & The Goal
★★ Adil says low angles with handheld shake give the real sports-ad feeling. — 00:22:07
★★ Short, punchy close-ups (the goal born from electric light, building up and exploding) make the ad look impressive and expensive. — 00:22:33 · also in: Make a Cinematic Ad End-to-End: Scenes 3-4: First Meeting & The Goal
★★ In the goal close-ups Seedance 2.0 made the goal forming out of electric lines look real, but the goal's explosion did not land. — 00:23:49 · Seedance 2.0 · also in: Make a Cinematic Ad End-to-End: Scenes 3-4: First Meeting & The Goal
★★★ For scenes with a lot of moving mechanics (two robots, a ball, feints) Adil runs many batches and cuts different phases of each generation together, which looks more organic. — 00:25:02 · rebatching
★★ Adil reorders the ad so the hero loses first, as in every sports film, saving the hero's goal for the end. — 00:25:31 · also in: Make a Cinematic Ad End-to-End: Scenes 5-6: The Duel & Round One Loss

00:26:19 — Scene six: the goals jump around the street, so a location scheme pins them in place

★★★ With no anchor for the goal's position, Seedance 2.0 put the goal in a different place in every generation and sometimes showed the goal input full screen as a keyframe. — 00:26:43 · Seedance 2.0
★★ For the location scheme Adil makes a high-angle wide shot of the location in GPT Image 2 and draws two dots by hand where the goals stand. — 00:27:31 · GPT Image 2
★★ From the marked image and the goal prop sheet, GPT Image 2 made a full production scheme with goal distance, street size, distance to the buildings and a compass. — 00:28:12 · GPT Image 2 · also in: Make a Cinematic Ad End-to-End: Scenes 5-6: The Duel & Round One Loss
★★ Adil declares the location scheme to Claude: the goals must appear exactly according to the scheme, and the goal shields are a reference, not a keyframe; the scheme is also added to Elements. — 00:28:51 · Claude, Elements · also in: Make a Cinematic Ad End-to-End: Scenes 5-6: The Duel & Round One Loss
★★★ With the location scheme the goals stay in place; Adil's tip: when words keep failing, show the model a picture, since "a map beats a paragraph". — 00:29:23 · also in: Make a Cinematic Ad End-to-End: Scenes 5-6: The Duel & Round One Loss

“A map beats a paragraph.”

00:29:59 — Scenes seven to twelve: the fall, the two-part comeback, round two, the slide tackle, the victory gag and the pack shot

★★ In the first takes of the robot's fall, Seedance 2.0 made the armour slice off the body in a second instead of breaking like metal. — 00:30:51 · Seedance 2.0
★★ Adil's physics note to Claude: when the robot hits the ground the armour breaks into heavy metal pieces, with no smooth breaks, like real metal snapping under pressure. — 00:31:21 · Claude
★★★ An object that appears only once in the video needs no asset of its own: the robot's watch was only mentioned in the prompt. — 00:31:46 · also in: Make a Cinematic Ad End-to-End: Scenes 7-8: The Fall & The Comeback
★★★ When a scene gets too packed, Adil splits it into smaller parts, one generation each, to give each moment room to breathe. — 00:32:19
★★ When the hero's reaction felt rushed (he saw the can and instantly smiled), Adil asked Claude to add air between the actions. — 00:33:13 · Claude · also in: Make a Cinematic Ad End-to-End: Scenes 7-8: The Fall & The Comeback
★★ In the second part of the comeback Seedance 2.0 left out the broken armour pieces of the previous part and had the can open by itself. — 00:34:34 · Seedance 2.0 · also in: Make a Cinematic Ad End-to-End: Scenes 7-8: The Fall & The Comeback
★★ For continuity Adil tells Claude the scattered armour pieces lie on the asphalt from the start and the hero opens the can himself, with no automatic hiss. — 00:35:17 · Claude
★★ For the rematch Adil builds on what worked: he asks Claude to take the scene 5 duel prompt and add more football feints. — 00:36:56 · Claude
★★★ To save credits Adil reuses takes across scenes: dribbling phases of the first duel that never made its edit go into the rematch, so nothing is wasted. — 00:38:06 · also in: Make a Cinematic Ad End-to-End: Scenes 9-10: Round Two & The Winning Goal
★★ For the winning goal Adil generates only the missing slide tackle and reuses the goal already made in scene 5, instead of generating the whole sequence. — 00:38:38 · also in: Make a Cinematic Ad End-to-End: Scenes 9-10: Round Two & The Winning Goal
★★ Adil lets Claude pitch the closing comic gag rather than writing the joke himself. — 00:40:08 · Claude · also in: Make a Cinematic Ad End-to-End: Scenes 11-12: Victory Gag & Pack Shot
★★ Adil calls the pack shot the scene that sells: it is the last one viewers see, and an impressive one is remembered. — 00:41:32 · pack shot · also in: Make a Cinematic Ad End-to-End: Scenes 11-12: Victory Gag & Pack Shot
★★ For the TOPUP name in the pack shot Adil asks for a slow camera pullback as the letters build up from pieces of metal locking together. — 00:42:30 · pack shot, Claude · also in: Make a Cinematic Ad End-to-End: Scenes 11-12: Victory Gag & Pack Shot

00:43:05 — The full commercial played

00:45:10 — Recap: iteration is the skill; swap the can for your product

★★★ Adil sums up that the final film is the best few seconds of many tries cut together: iteration is the skill. — 00:45:38 · also in: Make a Cinematic Ad End-to-End: Recap: Iteration Is the Skill

“Iteration is the skill.”
★ Adil says the workflow stays the same for any product and idea: swap the can and the robots for your own. — 00:45:38 · also in: Make a Cinematic Ad End-to-End: Recap: Iteration Is the Skill

In the description and comments

★★ Higgsfield: Asked by a viewer who says Claude Fable 5 got suspended how to connect to it, Higgsfield replies that the workflow works perfectly on Claude Opus 4.8 too. — creator comment · Claude Fable 5, Claude

Threads in this video

restate · How do you keep a character the same across shots? — For a character with his own likeness Adil calls Soul Cinema hands down the best model, keeping all facial features consistent: he selects Character and drops in 20 images of himself. (00:03:51)
refine · Which image model should be used for keyframes, edits, text and outfits? — Adil calls GPT Image 2 the best model right now for image editing. (00:03:07)
restate · Which image model should be used for keyframes, edits, text and outfits? — Adil calls the location the most important image: a sloppy or plasticky one ruins the videos whatever the prompt, since the video takes textures and lighting from it; Soul Cinema gives movie-still looks by default. (00:05:16)
restate · What does an AI clip, shot or film cost? — Adil says eight Soul Cinema images cost one credit, so five credits buy many outfit options, which makes the character sheet a step to experiment freely. (00:04:47)
refine · How do you avoid the AI (plastic) look? — Adil calls the location the most important image: a sloppy or plasticky one ruins the videos whatever the prompt, since the video takes textures and lighting from it; Soul Cinema gives movie-still looks by default. (00:05:16)
refine · How do you get a cinematic look in a generated image or shot? — In the Soul Cinema location prompt ("cinematic anamorphic lens, shallow depth of field, film grain"), the anamorphic lens gives the wide cinematic feel and film grain removes the too-clean digital look. (00:05:58)
restate · How do you keep a location, product or prop the same across shots? — Adil makes a product sheet of the TOPUP can in GPT Image 2 so that the video model knows the can from all angles and does not hallucinate it. (00:03:51)
refine · How do you keep a location, product or prop the same across shots? — With the location scheme the goals stay in place; Adil's tip: when words keep failing, show the model a picture, since "a map beats a paragraph". (00:29:23)
refine · Should prompts be short or long? — Adil says prompting is where most people fail: they overload their prompts with unnecessary information and waste time and credits. (00:10:48)
restate · Should sound and music be generated with the clip or added in the edit? — Adil's insight for video prompts: ask for no music and only environmental sound effects, because the music is added over the final video; the skill already includes this. (00:14:29)
restate · When a generation fails, should you generate more takes, fix the prompt or start over? — The rhythm of Adil's workflow: problems are not fixed by hand; you watch the takes, give notes and Claude rewrites the prompt. (00:18:46)
refine · When a generation fails, should you generate more takes, fix the prompt or start over? — For scenes with a lot of moving mechanics (two robots, a ball, feints) Adil runs many batches and cuts different phases of each generation together, which looks more organic. (00:25:02)
refine · How do you get precise motion (dance, action, choreography)? — Adil's physics note to Claude: when the robot hits the ground the armour breaks into heavy metal pieces, with no smooth breaks, like real metal snapping under pressure. (00:31:21)
restate · How do you direct a multi-shot sequence? — When a scene gets too packed, Adil splits it into smaller parts, one generation each, to give each moment room to breathe. (00:32:19)
refine · How do you direct a character's acting in a video prompt? — When the hero's reaction felt rushed (he saw the can and instantly smiled), Adil asked Claude to add air between the actions. (00:33:13)
refine · How do you spend fewer credits? — To save credits Adil reuses takes across scenes: dribbling phases of the first duel that never made its edit go into the rematch, so nothing is wasted. (00:38:06)
refine · How do you direct a multi-shot sequence? — When a generation misses, Adil stops giving Claude vague notes and directs shot by shot, describing what the camera sees in each shot (close-up, over-the-shoulder, low handheld angles) rather than what happens. (00:20:36)

Mentioned

Central: Claude · GPT Image 2 · Map technique · Seedance 2.0 · Seedance prompt Claude Skill (TOPUP ad) · Soul Cinema · TOPUP football ad

Also: Organizations: Jetix

Concepts: AI look · character sheet · native audio · pack shot · rebatching · style prefix

Places: New York

Products: Collab · Elements · Enhancer

Only in the description or credits: Claude Fable 5 · Claude Opus 4.8

Also centred on Map technique: How To Make Cinematic Football Ads With AI (Full Workflow) · 3-Step Workflow To Make Ultra-Realistic AI Ads · The 3-Step Realistic AI Ad Workflow: Scene 3 — the street, and the layout map trick · Make a Cinematic Ad End-to-End: Scenes 5-6: The Duel & Round One Loss · AI Got the Seating Wrong So I Drew a Map

Also centred on Seedance prompt Claude Skill (TOPUP ad): How To Make Cinematic Football Ads With AI (Full Workflow) · Make a Cinematic Ad End-to-End: Intro: One Website, One Cinematic Ad · Make a Cinematic Ad End-to-End: Stage 2: The Prompting Framework · Make a Cinematic Ad End-to-End: Recap: Iteration Is the Skill

Also centred on TOPUP football ad: How To Make Cinematic Football Ads With AI (Full Workflow) · Make a Cinematic Ad End-to-End: Intro: One Website, One Cinematic Ad · Make a Cinematic Ad End-to-End: Stage 2: The Prompting Framework · Make a Cinematic Ad End-to-End: Scenes 7-8: The Fall & The Comeback · Make a Cinematic Ad End-to-End: Scenes 11-12: Victory Gag & Pack Shot · Make a Cinematic Ad End-to-End: Scenes 3-4: First Meeting & The Goal

← Previous: Higgsfield AI + Minecraft?! 🧱🤖 · Next → Mastering AI Character Consistency in Video and Images · Soul ID