# AI Video Prompts: Full Prompts and Setups

> The full prompts and setups behind our AI video remixes. Copy one and make your own with Twistavo.

Canonical page: https://twistavo.com/prompts
Updated: 2026-11-18

## Candid smartphone photo prompts: 5 realistic phone shots on Z-Image and GPT Image 2.5

Made with Z-Image Turbo on 2026-11-18.

One vertical 9:16 image per prompt, text to image, no reference image. The words that make the photo look like a phone snapshot are the same in all five: "candid" or "casual phone photo", an ordinary place (laundrette, checkout, pavement), the real light of that place (fluorescent, overcast), real skin and cloth texture, and bystanders who react. Each prompt also says "original fictional" for every person. In Twistavo: open the AI Studio, pick Z-Image Turbo or GPT Image 2.5, paste the prompt, set 9:16 and the resolution shown, and generate. Z-Image Turbo costs a fraction of GPT Image 2.5 high (compare the cost under each example), so it suits trying many variations; GPT Image 2.5 follows long character descriptions closely.

Variations to try: the same prompt with "flash on, slightly overexposed" for a night snapshot, a lower camera height ("photo taken from hip height"), or the same character in three ordinary places for a recurring series. Every person on this page is fictional and was generated with AI with the prompt and settings shown. Results change from run to run.

- [See the full prompt for Candid smartphone photo prompts: 5 realistic phone shots on Z-Image and GPT Image 2.5](https://twistavo.com/prompts/candid-smartphone-photo)

## Character turnaround sheet prompts: 6 tested reference sheets, from 4 angles to a full character bible

Made with Seedream 4.5 on 2026-11-18.

One image per prompt. The two Seedream 4.5 sheets share one template: exactly four equal panels on a 3:2 canvas (front, three-quarter, side, head-and-shoulders), the same height and outfit in every panel, plain light-grey background, neutral pose, and a ban on text, logos and known characters. Only the character paragraph changes. The Soul Cinematic sheet uses two panels (face close-up and full length) on white, the layout a face-consistency reference needs. The dog sheet was made on GPT Image 2 with one reference image, used for style only. The two character bible sheets use almost the same layout prompt (title block, seven views, close-ups, expressions, poses and scenes on a dark 3:2 board) and take 2 reference images of the character: a close portrait and a full-body sheet. In Twistavo: open the AI Studio, pick the model, paste the prompt, set the aspect ratio, add a reference image if the example uses one, and generate. Then upload the sheet as the reference image of your next image or video.

What we would do next time: ask for panel numbers in the prompt and remove them afterwards, so a swapped panel is easy to spot; and generate the sheet at the aspect ratio of the panels you plan to crop. Variations to try: an outfit sheet with one character in three costumes, or a bible sheet that keeps only the turnaround and the expressions when the scene photos are not needed. Every character on this page is fictional and was generated with AI with the prompt and settings shown. Results change from run to run.

- [See the full prompt for Character turnaround sheet prompts: 6 tested reference sheets, from 4 angles to a full character bible](https://twistavo.com/prompts/character-turnaround-sheet)

## Faceless history video with AI: 3 shots, 2 stills, the exact prompts

Made with Seedance 2.0 and Seedream 5.0 Pro on 2026-11-18.

Order of the short: an establishing shot of the caravan for the hook, the market, then the library, with the voiceover running over all of them. Each shot is a separate Seedance 2.0 text-to-video clip, 9:16, 720p. The first pass made 6 clips from short prompts at the duration shown under each example; the second pass made 6 shorter clips from longer prompts that add a camera move and a symbol for each scene. The clips came with Seedance native sound (wind, market noise, candle crackle) and every prompt ends with "no music", so no music track was used; the clips on this page are muted. The two voiceover lines are each a little shorter than one first-pass clip, in French, made with a preset text-to-speech voice and a style tag that ends with "starts speaking immediately". The stills were not used as start frames: they are separate images of the same scenes, useful as a thumbnail and as a fallback when a clip fails. To assemble: put the voiceover on the timeline first, cut each clip to the length of its sentence, and keep the native ambience under the voice at a low level.

What we would change: the map shot of the second pass (a0daa5d1) showed a modern world map with the Americas, which does not fit a medieval story, so we left it out; ask for "a hand-drawn medieval map of the Mediterranean and the Indian Ocean" instead. Variations to try: the same seven scenes for another period (Silk Road, Hanseatic ports), a shorter cut for each sentence, as in the second pass, or a start frame from each still so the clip and the thumbnail match. Every clip and image on this page was generated with AI on August 5, 2026 with the prompt and settings shown. Results change from run to run.

- [See the full prompt for Faceless history video with AI: 3 shots, 2 stills, the exact prompts](https://twistavo.com/prompts/faceless-history-video)

## Kling 3.0 prompts: 3 stills brought to life with image to video

Made with Kling 3.0 Standard on 2026-11-18.

Two steps per clip. Step 1: Soul 2 text to image, 16:9, 1080p, one image, prompt enhancement off. Step 2: Kling 3.0 Standard image to video with that image as the start frame, 16:9, sound off, CFG scale 0.5, at the duration listed under each example, multi-shot off. The motion prompt describes only what moves and how the camera moves; the look comes from the still. In Twistavo: open the AI Studio, pick Kling 3.0 Standard, upload the still in the start-frame slot, paste the motion prompt, set that duration and 16:9, then generate. The cost under each example is the clip; the Soul 2 still is priced separately.

What works with Kling 3.0 image to video: one action plus one camera move per clip, in a single sentence; the subject from the still keeps the same face and clothes for the whole clip. Variations to try: a longer clip with a second action ("then sits down at a table"), sound on for street ambience, or Kling 3.0 Pro for finer detail at a higher price. All three stills and clips were generated with AI on October 5, 2026; the people in them are not real. Results change from run to run.

- [See the full prompt for Kling 3.0 prompts: 3 stills brought to life with image to video](https://twistavo.com/prompts/kling-3-image-to-video)

## Location reference sheet prompts: 3 empty places in three views each

Made with Nano Banana 2 Lite on 2026-11-18.

All three sheets use Nano Banana 2 Lite, text to image, aspect ratio 16:9 and one image per run; the output is 1376x768. The prompts share one structure: the word "Location reference sheet" first, then the layout (three views side by side, thin white gaps), then the place in one sentence of concrete materials and light, then one line per panel with the camera height and direction, and a closing ban on people, text and logos. Two sheets ran with thinking set to HIGH and one with MINIMAL.

Every place on this page is AI-generated from text: no photo of a real location was uploaded and no person appears. Two takes from the same series are not shown. A Champs-Elysees traffic sheet asked for a jam of ordinary cars and came back with readable maker badges on the cars and a city bus with an advert, which makes it unusable as a reference. A night sheet of the lit Eiffel Tower was left out for the lighting rights noted above. Sheets made in early October from frames of other people's videos are not on this page either. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026. Results change from run to run.

- [See the full prompt for Location reference sheet prompts: 3 empty places in three views each](https://twistavo.com/prompts/location-reference-sheet)

## The miniature film-set making-of: an AI video format, beat by beat

Made with Seedance 2.5 on 2026-11-18.

Every clip is one continuous Seedance 2.5 take at 480p, at the duration listed under it, 9:16, with the model's own sound on; nothing was cut or edited, and the clips here are muted for the web. The prompts split the take into five beats, each with its own second range written in the prompt: in the first two seconds the set is already alive and the director gives the signal (the hook: crew, camera and miniature are all on screen before anything happens), then the event starts, it hits the set, it spreads, and the take ends on the aftermath and the crew's reaction. Each clip also got two or three uploaded images that fix the start, the creature or the end state; they are not shown here. The storyboards were made first with GPT Image 2.5 as one 4k sheet of 16 vertical panels, to agree on the beats before paying for video. The sound in the original clips is the model's native audio (set noise, water, cries); no music was asked for.

Everything on this page is AI-generated: the crew, the studio and the miniatures do not exist, and no real film set was used. The format is our own take on the behind-the-scenes miniature idea; no third-party video was copied. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026. Results change from run to run.

- [See the full prompt for The miniature film-set making-of: an AI video format, beat by beat](https://twistavo.com/prompts/miniature-film-set-making-of-ai-video)

## Recurring AI characters for a series: one portrait, one 3-view sheet, three new heads

Made with GPT Image 2.5, Seedream 5.0 Pro and Nano Banana 2 Lite on 2026-11-18.

Order matters. The portrait fixes the face, the hair and the outfit; nothing else in the chain is text-only. The sheet prompt is a long reusable template (about 4,000 characters) that says to copy the person from the attached image and adds a single line at the end to complete the outfit below the waist, since a waist-up portrait never shows trousers or shoes. Only that last line changes between CLAUDE and BRIGITTE. The head prompts are a short edit template: keep the body, the suit, the pose and the light, replace only the head and hair. In Twistavo: open the AI Studio, generate the portrait from text, then pick the sheet model, upload the portrait as the reference image and paste the sheet prompt. For new heads, upload the sheet and paste the head prompt. Each character costs one portrait and one sheet; each head test costs one image.

Every person on this page is fictional and was generated with AI from the prompts shown; the chain starts from text, with no photo of a real person. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026. Results change from run to run.

- [See the full prompt for Recurring AI characters for a series: one portrait, one 3-view sheet, three new heads](https://twistavo.com/prompts/recurring-ai-character-cast)

## Seedance 2.5 character sheet to video: three one-take comedy clips with two recurring characters

Made with Seedance 2.5 on 2026-11-18.

The three prompts share the same opening block, word for word: the sheets are named @Image1 and @Image2, each character is described in one sentence (feathers, eye colour, clothes), and the prompt says the references control identity and clothes only, not poses or backdrop. The same block bans captions, logos, split screens and music, and asks for native sound effects without dialogue. Only the last paragraph, the gag itself, changes. The two sheets are the Seedream 4.5 turnarounds published on our character sheet page. The clips were made with sound on; the copies on this page are muted and compressed for the web. In Twistavo: open Create, choose Seedance 2.5, upload the two sheets as reference images, paste the prompt, set 9:16, the duration and the resolution listed under each clip, check the quote and generate.

Both characters are fictional and were generated with AI; the sheets are the ones on our character turnaround sheet page. The clips had model-generated sound; the prompts ask for no music and no dialogue, and the copies here are muted. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026. Results change from run to run.

- [See the full prompt for Seedance 2.5 character sheet to video: three one-take comedy clips with two recurring characters](https://twistavo.com/prompts/seedance-2-5-character-sheet-to-video)

## Seedance 2.5 prompts: 3 found-footage clips that look filmed on a phone

Made with Seedance 2.5 on 2026-11-18.

Seedance 2.5 text-to-video, 9:16, one continuous take, native audio on. The whale and cinema clips ran at 480p with the standard bitrate; the kayak clip ran at 720p with the high bitrate. The clips on this page are the original results, muted and compressed for the web: the originals carry the model's own sound (no music was asked for or added).

All three clips are AI-generated: the people, places and animals do not exist. Footage built to look like it was filmed on a phone should be labelled as AI when you publish it, and most platforms now ask for that label. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026; your quote can change with the settings you pick. Results change from run to run.

- [See the full prompt for Seedance 2.5 prompts: 3 found-footage clips that look filmed on a phone](https://twistavo.com/prompts/seedance-2-5-found-footage-prompts)

## Storyboard sheet prompts: 4x4 and 3x2 grids on GPT Image 2, Seedream 5.0 Pro and GPT Image 2.5

Made with GPT Image 2.5 on 2026-11-18.

One image per prompt, text to image, no reference image. What the three prompts share: the exact grid ("4x4 grid, 16 distinct panels", "strict 3-column by 2-row grid"), the reading order, one beat per panel written in order, one line that locks the character and the place for every panel, and a ban on captions, numbers and logos inside the panels. The longer the story, the longer the prompt: the 6-panel board uses about 3,900 characters and the 16-panel Pompeii board about 12,600, written in French. In Twistavo: open the AI Studio, pick the image model, paste the prompt, set the aspect ratio and resolution shown, and generate. To turn a board into video, crop a panel and use it as the start frame, or give the whole board to a video model as a reference image.

What we learned: a board costs far fewer credits than the clips it plans, so we fix the story on the board first. The models follow a grid size and a panel list well; they lose continuity on the details that change between panels (an object that appears once, a smoke column). Variations to try: the same story at 2x2 for a 4-beat short, panel numbers turned on for editing, or the same prompt on two models to compare. Every board on this page was generated with AI with the prompt and settings shown; the characters are fictional. Results change from run to run.

- [See the full prompt for Storyboard sheet prompts: 4x4 and 3x2 grids on GPT Image 2, Seedream 5.0 Pro and GPT Image 2.5](https://twistavo.com/prompts/storyboard-sheet)

## Talking animal sitcom with AI: Barbary macaques speaking Darija

Made with Gemini Omni Flash 1.1, Seedance 2.5 and Seedream 5.0 Pro on 2026-11-18.

Every clip is vertical 9:16. The kitchen board is one Seedream 5.0 Pro image from text only. The board was then cut into separate panel images, and those panels went in as reference images: five for Gemini Omni Flash 1.1, six for Seedance 2.5. Gemini wrote the voice itself from the dialogue in the prompt (native audio). For Seedance we gave a separate voice file as the audio reference and asked it to lip-sync to it. The sofa scene uses one image of the two characters as its only reference. The dialogue is in Arabic script inside English prompts; the gloss under each prompt is our translation and was not sent to the model. The clips on this page are muted and compressed for the web.

What went wrong: on September 29 we ran ten sofa takes with the same reference image and the same prompt opening, which says "Clothes plain; no emblem or visible writing". The reference image itself shows a football federation crest on the navy hoodie, and the model carried it over: the plain-clothes line gave a plain hoodie in only 1 of the 10 takes, the one shown above. In eight takes the hoodie had a sportswear logo, and in five of those also the crest. A ninth take gave the hoodie character a wristband with a printed badge. The lesson: clean the reference image before you generate, because a line in the prompt does not reliably remove what the reference shows. Then check frames before you pay for a second part. Every animal on this page is invented and every clip is AI-generated. The voice file of the Seedance example was made separately and is not priced here. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026. Results change from run to run.

- [See the full prompt for Talking animal sitcom with AI: Barbary macaques speaking Darija](https://twistavo.com/prompts/talking-animal-sitcom-ai-video)

## Talking dog talk-show video with AI: photo, set, voice and the episode

Made with Seedance 2.5 and GPT Image 2 on 2026-11-18.

Images with GPT Image 2 at quality high and 2k; video with Seedance 2.5 from one image, sound off. The voice line was generated separately as a 24 kHz WAV with Seed Audio from this text: "Hey, it's Bruno. Welcome to the show. First question: do I stare at the wall? No. I meditate. That wall owes me money." It was then given to Seedance 2.5, as an MP3 file, as the audio input of the talking clip. The talking clip comes back without a sound track (sound was off), so the voice file goes back under the picture in your editor. The pieces add up to one episode: the push-in opens the show, the talking clip carries the joke. The clips here are muted and compressed for the web.

Bruno is an invented dog: every image and clip on this page is AI-generated. A real pet photo works the same way as the input of step 5. The voice model (Seed Audio) is not priced in Twistavo's AI Studio quotes, so its cost is not listed; you can also record the line yourself. Credit costs are Twistavo's quotes for the same model and settings on November 18, 2026. Results change from run to run.

- [See the full prompt for Talking dog talk-show video with AI: photo, set, voice and the episode](https://twistavo.com/prompts/talking-dog-talk-show-ai-video)

## GPT Image 2.6 prompts: 5 tested examples

Made with GPT Image 2.6 on 2026-11-05.

GPT Image 2.6 text-to-image, one image per run unless noted. Each example lists its aspect ratio, quality and resolution. All 5 images were generated on November 5, 2026 and are shown unedited (resized for the web).

Every image on this page was generated with AI on November 5, 2026, with the prompt and settings shown. Results change from run to run: generate a few and keep the best one.

- [See the full prompt for GPT Image 2.6 prompts: 5 tested examples](https://twistavo.com/prompts/gpt-image-2-6)
