Prompting

Image-to-video (any model) · UGC / talking head

• Perfect the starting frame first; @maxxmalist puts it at 50%+ of creation time.
• A reference image beats a plain prompt for skin, lighting and framing.
• Keep the motion prompt physical and short: one action, one camera move.

Phone-camera selfie framing, slight handheld sway. She lifts the serum bottle into frame, glances at it, then back to camera and smiles before speaking. Natural window light from the left.

Sources: @maxxmalist

Wan 3.0 reference-to-video · Product / consistency

• Give each reference a job with @Image1/@Video1/@Audio1.
• Connect references in a fixed order; numbering follows input order.
• Up to 10 images, 5 videos, 5 audio clips.

@Image1 (the runner) laces up @Image2 (the shoe) on a wet city curb at dawn, matching the camera move in @Video1. Ambient traffic, rhythm from @Audio1.

Sources: ComfyUI docs

Kling / Veo / Seedance · Cinematic

• Write one shot brief: subject, action, camera, environment, timing.
• Test with the same first frame across models; count usable takes.
• Approve at 720/1080p, then render 4K.
• For exact dialogue, generate the voice separately and lip-sync.

Subject: lighthouse keeper, 60s, wool coat. Action: climbs the last stairs, pauses at the window. Camera: slow push-in, 35mm, low angle. Environment: storm at dusk, rain on glass. Timing: 8 s, pause at 5 s.

Sources: Pixraft comparison

Claude + Three.js (single HTML) · 3D explainer

• Ask for one HTML file with Three.js.
• List the outer parts, then 'the parts you never get to see'.
• Give one motion rule and the interactions (orbit, zoom, Explode button).

Build a mechanical watch in interactive 3D in one HTML file with Three.js: gold case, dial, hands, and behind it the movement you never get to see (mainspring, gear train, balance wheel, escapement). The balance wheel swings and the seconds tick. Drag to orbit, scroll to zoom, Explode button slides parts apart with labels.

Sources: Skillry gallery (@aichrislee pattern)

HyperFrames / Remotion motion · Motion graphics

• Name one typeface and its weights, and give the palette as hex codes.
• Forbid cuts if you want a one-shot morph ('no cuts, no fades; every scene is made from the previous one').
• Ask for music and SFX generated in code and synced to cuts.
• Keep the preview open and give notes by timestamp.

30-second looping showreel, 1080×1920, 60 fps. One line morphs through three worlds: graphite on cream paper (#EDE6D5), wire on blueprint (#1B3ED2), light on black (#07080A). Geist 900 for big words, 700 for labels. No cuts, no fades. Score and SFX generated in code, hits on every transform.

Sources: Skillry gallery · HyperFrames prompt guide

JS canvas animation · Animation / stick-figure / doodle

• Pick a named style (cut paper, crosshatch, riso, sketchbook, pixel, isometric).
• Ask for a beat table first and approve it before animation.
• Use seeded randomness and 'line boil on 2s' for a hand-drawn feel.

A 40-second sketchbook-style story of a paper crane crossing a city. Graphite plus one red accent. Story beats first as a numbered table; then 3 style frames; then animate. Lines boil on 2s. Piano score in code; silence before the final reveal.

Sources: animate skill · @kevin_t_ngo

Higgsfield Soul ID · Consistent characters

• Train on 20–80 varied photos, including a full-height shot.
• Call the character by name; don't re-attach references.
• Only fictional or consenting identities.

Generate an image using my Soul character Mara in an editorial fashion setting, studio lighting, neutral background, 85mm.

Sources: Higgsfield guide

ACE-Step 1.5 · Music

• Set duration, BPM, key and time signature explicitly so the edit grid is known in advance.
• Use repaint to fix one section instead of regenerating.
• Train a LoRA on a few of your own songs for a house sound.

Genre: cinematic synthwave. 96 BPM, A minor, 4/4, 30 seconds. Structure: 8-bar build, drop at 0:20, clean ending on beat. Instrumental.

Sources: ACE-Step 1.5

Product hero shot (Video generation)

Slow dolly-in on a matte black wireless speaker on a wet slate surface, soft rim light, shallow depth of field, 35mm, moody studio, 5 seconds.

Dialogue close-up (Video generation)

Close-up of a woman in her 30s speaking to camera in a sunlit kitchen, natural handheld micro-movement, 50mm, warm tones, subtle background bokeh.

Action tracking shot (Video generation)

Tracking shot following a cyclist through a narrow European street at golden hour, camera at hip height, motion blur on background.

Text in frame (Video generation)

Neon sign reading 'OPEN LATE' flickering on in a rainy alley at night, reflections on puddles, static locked-off shot.

Warm explainer VO (Voice, music & dubbing)

Voice direction: warm, conversational, mid-30s, medium pace, smile in the voice, slight pause before key numbers.

Avatar presenter script opener (Avatars & talking heads)

Hook in under 8 seconds: state the problem, the promise, and who it's for. Keep sentences under 12 words for clean lip-sync.