A Halloween spot where your product haunts the frame
Your agent builds the brand kit, turns your product into a flickering hologram and your presenter into a double exposure filled with a moonlit street, puts one line through a demon voice and cuts it all to the beat, rendered in GroundSlate on your Mac.
Make a 20 second Halloween spot in GroundSlate for <<BRAND>>, a <<PRODUCT>>, in <<PROJECT IN GROUNDSLATE>>, 1080x1920 at 30fps. 1. Build the brand kit first as an HTML kit page: a wordmark and its lockups, a named palette with hex values, a type pairing from fonts in my GroundSlate font library (list_fonts), a pattern drawn from the mark, and the motion pieces (a logo sting built from that pattern, a kinetic super, a transition). Save the page into GroundSlate as a kit with import_kit and render me its sheet with kit_detail before you make anything else. 2. With my own image and video models, make one master plate of the product (the product alone, true proportions, no lettering), a still of the product in a dark room at night, and a still of <<AVATAR>> against a plain light backdrop with hard side light. Move each still with an image to video model. Bring every take into the project with fetch_media and its provenance, check each with sample_frames, and reject the ones that fail with reject_generation. 3. Make a moonlit Halloween street still with no people and no lettering, and give it a slow camera move with apply_ai_effect depth_parallax. 4. List the looks with list_ai_effects and look-test before you render: preview_ai_effect double_exposure on the presenter's take with the street clip as the fill and a light paper, so a light-haired face keeps its own look; and preview_ai_effect hologram on the product take with particles at 0 and a colour from the kit. Apply the settings I approve with apply_ai_effect. 5. Speak one short line for the product in <<VOICE>>, with synthesize_speech on the Mac or my voice model through provider_request, slow it, and treat it with apply_voice_effect demon. Keep the intensity low enough that every word still reads, and check it by transcribing the treated take. 6. Make a dark music bed at 115 BPM or faster with generate_music, a low boom and a projector flicker with generate_sound_effect, and get the beat map with audio_beats. 7. Build the spot as one HTML composition and render it with render_html, in parts if it is long, and join them with render_concat: the presenter's double exposure first, the product switching between real and projection on the beats, the line placed exactly where the script puts it with the projection stuttering on its words, the kit's supers, never on the moving product, and the kit's logo sting with no smoke, dust or particles. Nothing else on screen. 8. Mix it, check the loudness, and give me a 2 fps contact sheet, the kit sheet and a list of every paid request with its cost. Tell me what did not resolve.
Everything in <<double brackets>> is yours: the brand, the product, the project, the avatar and the voice. Fill them in before you paste.
- The kit comes first, then one master plate, then the shots. That order keeps the product identical in every frame.
- Look-test both effects on one frame before you render a clip. A dark paper inverts a light-haired presenter's double exposure, and the hologram's rising dust is on by default.
- The flicker is the cut: the same take as shot and as a projection, switched on the beat.
- Our run: the AI in the finished spot cost $1.02 at list price (the stills and three video clips from our own models); every request, including the kit's extra cut-out and an unused voice take, came to $1.08. The effects, the kit, the voice treatment, the music, the edit and the sting ran on the Mac.