Film School

Lesson 1: How AI video tools work

Tools like Seedance inside Higgsfield turn a written prompt (and optional reference images) into a short video clip. Think of each generation as one short scene with its own sound.

• One generation is short. Seedance clips run up to about 15 seconds, so longer videos are built from several clips edited together.
• Every prompt stands alone. The model doesn't remember your last clip, so repeat the important details each time.
• Choose the aspect ratio first. 16:9 for YouTube and film, 9:16 for TikTok, Reels and Shorts, 1:1 for feeds. It changes how you frame every shot.
• Keep prompts focused. Clear, specific prompts beat long, crowded ones. Tool limits change over time, so check the current limits in Higgsfield.

Lesson 2: Style and purpose

Before writing anything, decide two things: the look (cinematic, photorealistic, anime, 3D animation, vintage film) and the job of the video (short film scene, ad, music video, social clip). Every other choice follows from these.

Write the look as one clear sentence at the top of your prompt, for example: Cinematic, rich contrast, shot on a cinema camera with anamorphic lenses, shallow depth of field.

Lesson 3: Shots and structure

Short AI clips work best when they're broken into numbered shots, in the order things happen.

• Plan about 2–3 seconds per shot, so a 15-second clip holds roughly 4–6 shots.
• For each shot, write the timing, the shot size (wide, medium, close-up, POV, low angle) and the action.
• Put the most important action first in each shot.
• One continuous take is a great option for calm or dreamy scenes.

Lesson 4: Camera language

The camera is where most beginner prompts go wrong. The rule that fixes it: one camera move per shot.

• Useful moves: slow push in, pull back, dolly left/right, crane up, orbit, tracking shot, whip pan, handheld follow, static.
• Pick one camera look for the whole clip (for example a cinema camera with anamorphic lenses, or a phone vlog look) and keep it.
• Be specific. "Fast tracking shot beside the runner" beats "dynamic camera".

Lesson 5: Characters that stay consistent

Faces and outfits can drift between clips. To keep a character recognizable:

• Describe each character once, fully: age, build, face, hair, clothing colors, and one distinctive detail (a red scarf, a silver ring).
• Give them a name and use only that name afterwards.
• Upload reference images in Higgsfield and tag them in the prompt so every clip uses the same face.
• Avoid real celebrities and copyrighted characters; create original look-alikes instead.

Lesson 6: Lighting, time of day and mood

State the time of day and make the light match it. Describe light physically: low warm sun and long shadows, cool moonlight with orange streetlamps, soft window light. Mood words work best when the lighting supports them, so never mix "bright noon" with "dark and moody".

Lesson 7: Describe motion physically

AI follows what it can picture. Instead of abstract words like "energetic" or "epic", write what the viewer would actually see: her hair lifts in the wind, sparks burst as the blades meet, coffee splashes in slow motion.

Lesson 8: Sound, music and dialogue

Seedance generates sound with the picture, so direct it:

• If nobody should talk, write "No dialogue." Otherwise the model may invent speech.
• Put spoken lines in quotes inside the shot where they happen, with the speaker's name. Keep them short, around 2–3 words per second.
• End with one sound line: music style, sound effects, or ambience.
• For longer films, ask for ambient sound only and add one soundtrack in your editor so the music doesn't change at every cut.

Lesson 9: Making longer films

A 5-minute film is about 20 clips of 15 seconds. Plan before you generate:

• Write a story bible: the look sentence, character descriptions and locations, copied word for word into every clip.
• Make a part map: what happens in each 15-second part and the exact image it ends on.
• Start each part where the last one ended, and change the shot size across the cut to hide small differences.
• Add music, voiceover and one color grade in your editor at the end.

Lesson 10: Workflow: test, compare, improve

Generate cheaply first, then polish:

• Test at lower resolution, then rerun only the winner in high quality.
• Make 2–3 versions of a scene that differ in one thing (camera, lighting, or shot breakdown) and keep the best.
• When a result is wrong, change one thing at a time so you know what fixed it.

Here is a complete example prompt built with these lessons:

Cinematic multi-shot battle sequence, 5 shots, 15 seconds, 16:9.

Epic heroic battle at golden hour in a ruined ancient city of red laterite walls and toppled bronze statues. Low warm sun, long shadows, glowing dust and drifting smoke. Shot on a cinema camera with anamorphic lenses, shallow depth of field.

Characters: Adé, a muscular warrior in his 30s with a shaved head, cowrie-studded brown leather armour, a red-and-white beaded sash, holding a long iron sword.

Shot 1 (0–3s, low angle close-up, slow push in): Adé stands on the fallen head of a bronze statue, sun blazing behind him. Adé shouts: "For our ancestors!"
Shot 2 (3–6s, high angle wide, crane up): The camera rises to reveal a bronze army filling the street.
Shot 3 (6–9s, tracking shot, slow motion): Adé leaps into the army, sash flying, and lands hard.
Shot 4 (9–12s, medium shot, fast orbit): He fights three soldiers at once; sparks spray in the golden light.
Shot 5 (12–15s, low angle wide, pull back): Adé stands among shattered bronze, sword raised.

Music: talking drums building into an epic orchestra. Sound effects: clanging bronze, sword impacts, wind. No other dialogue.