The Complete AI Filmmaking Masterclass Everyone’s Missing (Full Tutorial)
2026-10-24 · case-study tutorial · Adil · recording, 21:26 · watch on YouTube
Topic: Video generation & shot direction (also: Pre-production & story, Assets, references & consistency, Prompting & AI assistants, Post-production & finishing, Cost, credits & plans) · mode: Tutorial · tags: prompts shared, fiction, one creator's experience
Adil breaks down the action short he made in two days about Adam, a half-vampire werewolf hunter, with Claude, Seedance 2.5 and Cinema Studio: texture-fixed characters, video elements with recorded voices, anamorphic locations, 480p drafts, fight and acting skills, blocking references, new angles from one take and Extend.
What is said
The parts of the video and what is said in each (★ — how much it matters to the video).
00:00:00 — Adam, the film's hero, starts hosting the video until Adil takes over
00:00:37 — What the breakdown covers; a Cinema Studio project with a folder per scene; skills for scene structure, fights and acting
- ★★ Adil made his action short film in two days with Higgsfield and shows how he choreographed the fights, directed the acting and the camera, with his exact prompts, skills and steps. — 00:00:37 · Higgsfield AI, Adil
- ★★ Adil starts with a Cinema Studio project: one folder per scene with subfolders for characters, locations and props, since a little setup saves a lot of scrolling later. — 00:00:37 · Cinema Studio, Adil
00:01:21 — Designing Adam from Adil's photo in GPT Image 2.5, then fixing skin, hair and fabric with Seedream references
- ★★ For the hero Adam, a white-haired half-vampire who hunts werewolves, Adil gives Claude his own photo and a few notes and asks for two versions, which Claude expands into detailed prompts. — 00:01:21 · Claude, Adil
- ★★★ Adil usually makes his characters in Soul, but Soul does not support image-to-image, so to build a character from his own photo he uses GPT Image 2.5. — 00:01:21 · Soul, GPT Image 2.5, Adil
- ★★ Animated, Adil's GPT Image 2.5 version of Adam showed dirty plastic skin, dead-looking hair and many GPT artifacts moving around, which he says looks cheap. — 00:01:50 · GPT Image 2.5, Adil · also in: One AI Take, Four New Camera Angles
- ★★★ To clean up a character, Adil gives Seedream the model sheet plus reference shots for hair, skin, fabric and background, and says what to take from each image and what to keep, like texturing a 3D character. — 00:01:50 · Seedream, character sheet, Adil · also in: One AI Take, Four New Camera Angles
00:02:33 — Casting Victor with AI Cast and Ron in Soul Cinema
- ★★ Side by side, Adil finds the Seedream version of Adam much better: pores instead of patches, and hair that moves like real hair. — 00:02:33 · Seedream, Adil · also in: One AI Take, Four New Camera Angles
- ★★ For the antagonist Victor, Adil uses Cast: set budget, era and archetype, fine-tune the features and generate a quick batch, a shortcut when you know the type but not the exact look. — 00:02:33 · Cast, Adil
- ★★★ Adil generates many options for the heavyset gangster boss Ron in Soul Cinema, which he calls the cheapest yet the most creative model. — 00:03:21 · Soul Cinema, Adil
00:03:31 — Image element versus video element; simple voice recordings for each character
- ★★★ Adil makes a video element for each character before building scenes: an image shows only the look, a video also shows how the face moves and sounds, and the acting feels more natural. — 00:03:44 · Elements, Adil
- ★★ For each character's video element Adil uses a simple voice recording; he says a human tone makes the AI voices feel more alive without copying anyone's voice exactly. — 00:04:07 · Elements, Adil
- ★★ Adil recorded Adam's voice himself and asked friends to record the voices of Ron and Victor. — 00:04:07 · Adil
- ★★★ To give a character a recorded voice, Adil makes a video element: select the character generation, create an element, choose video, upload the audio, add a short prompt for how the character should move and speak, and generate. — 00:04:07 · Elements, Adil
- ★★★ For the alley location Adil asks for a three-quarter view for depth and a 24mm anamorphic look (wide framing, softer edges, oval bokeh); Claude expands it into a prompt he generates in Soul Cinema. — 00:04:40 · Claude, Soul Cinema, Adil
00:04:44 — Locations at a three-quarter view with 24mm and 35mm anamorphic looks
- ★★★ Adil uses a 35mm anamorphic look for a tighter frame of Ron's office; in motion, the softer background carried over from the anamorphic reference makes Adam stand out more. — 00:05:09 · Adil
- ★★ To avoid typing all the lens details every time, Adil uses an anamorphic skill in Claude. — 00:05:09 · Claude, Adil
00:05:42 — Watch: the phone call
00:06:07 — Scene prompts with elements, the Higgsfield prompt skill and Claude
- ★★★ Adil names and saves each asset as an element in Higgsfield and gives Claude the same file and name; Claude writes that tag into the prompt, so Higgsfield attaches the right element on paste. — 00:06:07 · Elements, Claude, Higgsfield AI, Adil
- ★★ Adil attaches the Higgsfield prompt skill, a set of instructions for Claude, to his scene brief, asks Claude to build a conflict, then asks for shots, camera moves and dialogue. — 00:06:07 · Claude, Adil
- ★★★ Adil says the Higgsfield prompt skill still needs direction: he spells out how Ron's anger builds, which lines he shouts and where the camera goes, since a clearer scene gives Claude a more specific prompt. — 00:06:07 · Claude, Adil
- ★★ The Seedance prompt Claude returns with the Higgsfield prompt skill lists the named elements, then optics for framing, camera for angles and movement, action timing second by second, and lighting, sound and acting. — 00:06:07 · Seedance 2.5, Claude, Seedance
00:06:56 — 480p drafts, take selection and a random Dutch angle; watch: the nightclub
- ★★ Adil generates the phone-call scene in Seedance 2.5 at 30 seconds and 21:9, running a few batches for different options. — 00:06:56 · Seedance 2.5, Adil
- ★★★ Adil's credit-saving tip: run drafts at 480p, where only framing and who stands where are checked, then recreate just the winning takes at 1080p. — 00:06:56 · credit, Adil
- ★★ In the phone-call batches, the first take gave Adam's starting shot with the camera rising behind him at a three-quarter angle; the others put Adam behind the booth or in a flat frontal shot. — 00:06:56
- ★★★ Adil says batching matters because the same prompt gives slightly different results: a random take gave him a low, tilted Dutch angle he never asked for, and he kept it. — 00:07:29 · rebatching, Adil
- ★ Adil suggests posting a shot with its prompt in the Higgsfield Discord to get feedback from the community. — 00:07:29 · Discord, Adil
00:08:48 — Nightclub location, crowd and extras with jobs, guards
- ★★ Adil made the nightclub location in Soul Cinema and builds the club scene in three parts, entrance, party and fight, then cuts the best shots together. — 00:08:48 · Soul Cinema, Adil
- ★★ When the first top-down take of Adam's club entrance made the crowd move oddly, Adil asked for a high side angle and a crowd moving naturally: jumping, landing, making room for Adam. — 00:08:48 · Adil
-
★★★ Adil's rule for crowds: every extra needs a job; different actions, such as a woman dancing towards the camera and two men enjoying drinks, make the crowd feel alive. — 00:09:15 · Adil
“every extra needs a job”
-
★★ When the club guards moved in sync and the baton looked fake, Adil gave each guard separate reactions and toned down the baton's electrical effect. — 00:09:52 · Adil
- ★★ Adil's first club-fight takes without the fight skill had soft punches and a stiff Adam. — 00:09:52 · Adil
- ★★★ For sharper strikes and reactions Adil uses a fight skill packed with real fight choreography and writes the fight as a list: who is where, what Adam does, how each guard goes down. — 00:09:52 · Adil
00:10:11 — Fight choreography with the fight skill; watch: the corridor fight
- ★★ In Adil's club fight medium shots show the move and close-ups make the contact feel real; he says the fight skill holds fight physics really well. — 00:10:28 · Adil
00:11:09 — Yakuza extras as one group; four takes combined with cuts hidden on movement
- ★★★ Adil generates the Yakuza crew of the corridor fight together as a group, so one reference asset serves all the extras instead of one per character. — 00:11:09 · Adil
- ★★ For the corridor fight, with the camera shaking and whipping inside the action, Adil uses the same fight skill but spells out where the foot goes, where it lands and which way the guards fall. — 00:11:09 · Adil
- ★★★ Adil built the corridor fight from four takes with every cut on movement, so the seams do not show; a whip pan's fast sideways movement gives a place to hide a cut. — 00:11:42 · Adil
00:12:13 — Re-lighting the office and adding its second side; two rules for dialogue
- ★★★ When the first batch of Victor's office came out cold and blue, Adil liked the room but not the mood, so instead of re-rolling he re-lit the image. — 00:12:13 · Adil
- ★★★ For a dialogue cut back and forth, Adil adds the office's second side, facing the door with the hallway behind; with the sofa behind Victor and the door behind Adam the model always knows which way it looks. — 00:12:13 · Adil
- ★★★ Adil's first rule for dialogue scenes: put the lines into the prompt word for word, or the model rewrites the script. — 00:12:13 · Adil
- ★★★ Adil's second rule for dialogue scenes: do not plan every shot; give the model camera rules and let the skill build the shot list. — 00:12:13 · Adil
- ★★ Adil wants the office dialogue as a duel lost by whoever loses his cool first; in the first take both men stay calm and a glass ruins the shot, so he has Claude revise the prompt with the acting skill. — 00:12:52 · Claude, Adil
00:13:07 — Dialogue and acting with the Higgsfield Acting Skill
- ★★★ The Higgsfield Acting Skill builds a performance around what each character is trying to get from the other person; Adil points to the difference it made in a test with Ron and Adam. — 00:13:20 · Adil
- ★★ With the Higgsfield Acting Skill on, Adil tells it who Adam and Victor are and what each wants, plus three fixes: a shorter line for Adam, no glass and room for pauses. — 00:13:20 · Adil
- ★★ Adil's next three fixes to Claude: drop the dust and splinters of the door kick, move around Adam to stay on his face during the line, and lock the room with the door behind Adam and the sofa behind Victor. — 00:13:50 · Claude, Adil
- ★★ Adil built the finished office dialogue from six generations, keeping the useful moments from different batches. — 00:13:50 · Adil
- ★★★ Adil's editing tip: repeat the last line of the previous generation in the next one, which gives an overlap for choosing where to cut from one generation to the next. — 00:13:50 · Adil
00:14:39 — The bullet shot with references and an explicit order
- ★★ In Adil's shooting sequence the bullet looked like a metal ball, kept changing shape and hole, and in one take went back into Victor's head. — 00:14:39 · Adil
- ★★★ Adil fixes the bullet shot with two references, the used bullet he wants and a photo of Victor with a wound of the right size, and an explicit order: the bullet comes out, falls, and the wound stays visible. — 00:14:39 · Adil
00:15:19 — Shrinking the fight stage; a blocking reference from a room tour
- ★★★ Two sides of a room work for dialogue but confused the model in Adil's fight, so he shrank the stage: everything happens within two meters of the red sofa. — 00:15:19 · Adil
- ★★★ Adil's trick: turn one location image into a short room tour in Seedance, which gives new views of the same space from different sides. — 00:15:19 · Seedance, Adil
- ★★★ Adil marks the start positions on the overhead frame of the room tour and attaches it in Claude and Seedance as a blocking reference for starting positions only, with the camera following the prompt. — 00:15:19 · Claude, Seedance, Adil
00:16:04 — New angles from one video reference; Extend in Cinema Studio
- ★★★ For wider angles of a fight, Adil gives Claude the previous take, asks for the same choreography from other camera positions, and feeds that prompt with the take as video reference into Higgsfield. — 00:16:04 · Claude, Higgsfield AI, Adil
- ★★★ When a take ends before the move is finished, Adil uses Extend in Cinema Studio, useful when he likes the shot and needs only a few more seconds to finish it. — 00:16:04 · AI Video Extender, Cinema Studio, Adil
- ★ The last scene of Adil's film was a single prompt, which is in the blog post with the others. — 00:16:44 · Adil
00:17:09 — Watch: the full short film
00:21:05 — Share your film in the Discord
- ★ Adil invites viewers to try the workflow in their own films and share them in the Higgsfield Discord, saying he would love to break down some of them in a future episode. — 00:21:05 · Discord, Adil
Threads in this video
- refine · Which image model should be used for keyframes, edits, text and outfits? — Adil usually makes his characters in Soul, but Soul does not support image-to-image, so to build a character from his own photo he uses GPT Image 2.5. (00:01:21)
- restate · Which image model should be used for keyframes, edits, text and outfits? — Adil generates many options for the heavyset gangster boss Ron in Soul Cinema, which he calls the cheapest yet the most creative model. (00:03:21)
- refine · How do you avoid the AI (plastic) look? — To clean up a character, Adil gives Seedream the model sheet plus reference shots for hair, skin, fabric and background, and says what to take from each image and what to keep, like texturing a 3D character. (00:01:50)
- refine · How do you keep a character the same across shots? — Adil makes a video element for each character before building scenes: an image shows only the look, a video also shows how the face moves and sounds, and the acting feels more natural. (00:03:44)
- refine · How do you keep a character's voice the same across shots? — For each character's video element Adil uses a simple voice recording; he says a human tone makes the AI voices feel more alive without copying anyone's voice exactly. (00:04:07)
- restate · How do you choose the location image for a generated scene? — For the alley location Adil asks for a three-quarter view for depth and a 24mm anamorphic look (wide framing, softer edges, oval bokeh); Claude expands it into a prompt he generates in Soul Cinema. (00:04:40)
- refine · How do you get a cinematic look in a generated image or shot? — Adil uses a 35mm anamorphic look for a tighter frame of Ron's office; in motion, the softer background carried over from the anamorphic reference makes Adam stand out more. (00:05:09)
- restate · How do you write a prompt that a video model follows in every detail? — Adil names and saves each asset as an element in Higgsfield and gives Claude the same file and name; Claude writes that tag into the prompt, so Higgsfield attaches the right element on paste. (00:06:07)
- refine · How do you write a prompt that a video model follows in every detail? — Adil's first rule for dialogue scenes: put the lines into the prompt word for word, or the model rewrites the script. (00:12:13)
- restate · How should a video prompt be structured? — The Seedance prompt Claude returns with the Higgsfield prompt skill lists the named elements, then optics for framing, camera for angles and movement, action timing second by second, and lighting, sound and acting. (00:06:07)
- refine · How do you spend fewer credits? — Adil's credit-saving tip: run drafts at 480p, where only framing and who stands where are checked, then recreate just the winning takes at 1080p. (00:06:56)
- restate · How do you draft and lock a shot before rendering the final? — Adil's credit-saving tip: run drafts at 480p, where only framing and who stands where are checked, then recreate just the winning takes at 1080p. (00:06:56)
- refine · When a generation fails, should you generate more takes, fix the prompt or start over? — Adil says batching matters because the same prompt gives slightly different results: a random take gave him a low, tilted Dutch angle he never asked for, and he kept it. (00:07:29)
- refine · How do you make a large crowd or army in a generated shot? — Adil's rule for crowds: every extra needs a job; different actions, such as a woman dancing towards the camera and two men enjoying drinks, make the crowd feel alive. (00:09:15)
- restate · How do you make a large crowd or army in a generated shot? — Adil generates the Yakuza crew of the corridor fight together as a group, so one reference asset serves all the extras instead of one per character. (00:11:09)
- refine · How do you get precise motion (dance, action, choreography)? — For sharper strikes and reactions Adil uses a fight skill packed with real fight choreography and writes the fight as a list: who is where, what Adam does, how each guard goes down. (00:09:52)
- refine · How do you make a transition between two shots? — Adil built the corridor fight from four takes with every cut on movement, so the seams do not show; a whip pan's fast sideways movement gives a place to hide a cut. (00:11:42)
- restate · How do you change the lighting of an existing image? — When the first batch of Victor's office came out cold and blue, Adil liked the room but not the mood, so instead of re-rolling he re-lit the image. (00:12:13)
- restate · How do you control where the camera is in a known set? — For a dialogue cut back and forth, Adil adds the office's second side, facing the door with the hallway behind; with the sofa behind Victor and the door behind Adam the model always knows which way it looks. (00:12:13)
- refine · How do you control where the camera is in a known set? — Adil marks the start positions on the overhead frame of the room tour and attaches it in Claude and Seedance as a blocking reference for starting positions only, with the camera following the prompt. (00:15:19)
- refine · How do you direct a multi-shot sequence? — Adil's second rule for dialogue scenes: do not plan every shot; give the model camera rules and let the skill build the shot list. (00:12:13)
- refine · How do you direct a character's acting in a video prompt? — The Higgsfield Acting Skill builds a performance around what each character is trying to get from the other person; Adil points to the difference it made in a test with Ron and Adam. (00:13:20)
- refine · How do you keep a location, product or prop the same across shots? — Adil fixes the bullet shot with two references, the used bullet he wants and a photo of Victor with a wound of the right size, and an explicit order: the bullet comes out, falls, and the wound stays visible. (00:14:39)
- restate · How do you change the camera angle of an existing video? — For wider angles of a fight, Adil gives Claude the previous take, asks for the same choreography from other camera positions, and feeds that prompt with the take as video reference into Higgsfield. (00:16:04)
- refine · How do you make a clip longer? — When a take ends before the move is finished, Adil uses Extend in Cinema Studio, useful when he likes the shot and needs only a few more seconds to finish it. (00:16:04)
- refine · How do you use your own voice in a generated video? — To give a character a recorded voice, Adil makes a video element: select the character generation, create an element, choose video, upload the audio, add a short prompt for how the character should move and speak, and generate. (00:04:07)
Mentioned
Central: Adam · Adil's half-vampire action short film · Cinema Studio · Claude · Elements · Fight skill · Higgsfield Acting Skill · Higgsfield prompt skill · Seedance 2.5 · Soul Cinema
Also: People: Adil
Organizations: Discord
Concepts: character sheet · credit · dutch angle · rebatching · shot list · Skill · whip pan
Characters: Ron · Victor
Products: AI Video Extender · Anamorphic skill · Cast · GPT Image 2.5 · Seedance · Seedream · Soul
Also centred on Adam: One AI Take, Four New Camera Angles
Replied to by
← Previous: This AI short drama got 60M views and doesn't look like AI · Next → This is why your AI locations never match