How to Write Prompts for Raz-2.0

1. Three modes: how your photo is used decides what you write

What is different from Raz-1.0: any length from 5 to 30 seconds, sound, no effect options, a 2,000-character prompt limit, and three generation modes. Each mode gives a different answer to "what happens to the photo I upload", and the prompt changes with it:

2. Four rules that apply everywhere

2.1 Tell it as a short story, not a list of keywords.

Write it the way you would tell someone what happens: who does what, then what happens next. A list of keywords gives the model nothing to follow.

woman, nude, bed, missionary, fast, 4k, photorealistic

He walks up to the woman and kisses her. Still kissing, he pulls her clothes off, and pulls off his sweatpants.

2.2 Write the shortest version first, then add only what is missing.

Start with one or two sentences of plain story — who does what, in order. Leave out camera, lighting, sound and style unless the result is missing something, then add one thing at a time. The "fix-it" lines in section 8 go in only when you actually hit that problem.

2.3 Describe the movement and how it feels, not numbers.

Say how the movement looks and feels — fast, deep, slow, rough, and what the contact does to the body: pressure, grip, skin on skin, flesh giving on impact. Numbers like "four strokes per second" don't turn into motion.

fast thrusting

sucks his cock fast and deep

deep insertion, tight grip, natural skin-on-skin friction, soft-body deformation on impact

four strokes per second

2.4 Write in your own language — except Arabic and Italian.

Write the whole prompt in one language; your own is fine. For Arabic and Italian, write in English. If the results keep coming out wrong, try English — it works best.

3. Image mode: movement + camera

• The photo already sets the person, the setting and the style — write only what moves. If you want the camera to do something, say it: the camera pushes in, the camera pans left, or fixed camera to keep it still.
• A person who is not in the photo: describe their looks and clothes the first time they appear — including what they wear below the waist. Whatever you leave out, the model decides for you, and clothes you mention later may appear out of nowhere.
• Write undressing as an action in progress (pulls her clothes off). The first frame is your photo, where she is dressed — a statement like "she is naked" contradicts it.

Template "Long Video" (first frame: a woman in a red dress; 15 seconds):

A muscular bald Black man walks into frame, bare-chested in grey sweatpants. He walks up to the woman and kisses her. Still kissing, he pulls her clothes off, and pulls off his sweatpants. She kisses down his chest, crouches in front of him and sucks his cock fast and deep. Then she pulls it out, holds it aimed at her face, and thick white cum shoots from the tip in arcs onto her forehead, nose and lips — from the cock in her hand, not from above or off-screen. She moans and says: "Oh god… yes…"

Notice the shape: four stages in time order (walks in and kisses → undresses → oral → cum), each as bare actions. For the cum shot, say where it starts and where it lands, and rule out from above or off-screen — without that it tends to look like it falls from the top of the frame.

End frame (optional). Add a second photo and the video ends on it; the prompt then describes what happens between the first frame and the end frame.

Template "First & Last Frame" (first frame: dressed, sitting; end frame: mid-penetration; 7 seconds):

She lies back on the bed pulling her clothes off, and he kneels between her legs and pushes inside — all in the first second and a half, smooth, no jumps. The rest is fast thrusting. She moans loudly and says: "Oh god… yes…"

Put visual details in the photos, not in the words — when the end frame shows something, the video follows the photo. If the move between the two photos jumps instead of flowing, give it a second or two more.

4. Reference mode: @photo + action + dialogue

• Refer to photos as @Image1, @Image2, numbered in upload order, and say who is who at the start. If you delete a photo in the middle, the ones after it are renumbered — check your prompt.
• Mention every photo by number, and you can @ it again at each action (@Image2 walks up to her, @Image1 kneels) — that is how you control which photo does what. A photo you never mention may be ignored, and you can't tell from the result. The one exception: when the photos already say everything, you can skip the numbers — the shortest example below does exactly that.
• You don't have to type the numbers: type @ in the prompt box, or tap the @ button in its corner, and pick a photo to insert @ImageN.

The shortest example — template "Couple from 2 Photos" (@Image1: a woman; @Image2: a man; 5 seconds):

Two people having sex.

The photos already say everything about the people, so one sentence is enough. To keep the room from one of the photos, write in the room from @Image1.

Two ways to write nudity, depending on whether you want to see the undressing:

• You want to see it → write it as an action in progress (he pulls her clothes off).
• You don't → don't write "undress" at all. State the condition: naked from the very first frame, wearing nothing, and clothing must never appear. Describe it as a state, not as something that has just happened.

The Reference version of "Long Video" tells the same story as the Image version, with four changes: it opens by saying who is @Image1 and who is @Image2; the setting is written as in the room from @Image1; the man's looks come from @Image2, so they are not described; and since his clothes are no longer set by text, "pulls off his sweatpants" becomes "takes his own off".

Template "Long Video", Reference version (@Image1: a woman; @Image2: a man; 15 seconds):

@Image1 is the woman, @Image2 is the man, in the room from @Image1. @Image2 walks up to her and kisses her. Still kissing, he pulls her clothes off, and takes his own off. She kisses down his chest, crouches in front of him and sucks his cock fast and deep. Then she pulls it out, holds it aimed at her face, and thick white cum shoots from the tip in arcs onto her forehead, nose and lips — from the cock in her hand, not from above or off-screen. She moans and says: "Oh god… yes…"

5. Text mode: person + setting + movement

With no photo, everything comes from the text. The basic shape is who + where + doing what, still as a skeleton:

Template "No Photo Needed" (5 seconds):

A dim bedroom. A naked woman in her twenties with long dark curly hair lies back on the bed, legs apart; a naked man kneels between them and fucks her fast. She moans loudly and says: "Oh god… yes… don't stop…"

One line for the setting → a few words for the person (age, hair, build) → position and action → sound. When the basic version works and you want more control, add one of these at a time:

• Light and camera: the light source and mood (warm afternoon light through the blinds), the framing (close-up, medium shot, low angle), and how the camera moves (slow push-in, fixed camera).
• Style: one phrase for the look (photorealistic, film grain, soft focus). Keep it to one — mixing styles is what turns a video cartoonish (see section 8).
• Sound: see section 6.

6. Sound: voice + effects + music

Videos come with sound by default; turning it off doesn't make it cheaper — the switch only says whether you want sound. Anything you don't describe, the model fills in on its own. To control it, write it:

• Voice: write says: "…" and the model says exactly that, in whatever language you write it. To shape it, add the mood, pace and tone after the line: says: "Oh god… yes…", breathless, slow, a low whisper. If you want no dialogue at all, write no dialogue.
• Sound effects: name what makes the sound and where: wet skin slapping, the bed creaking, a quiet room.
• Background music: if you say nothing, the model adds its own. To choose it, name a style (slow, sensual background music); to have none, say so:

No background music.

7. Duration and shots

Anything from 5 to 30 seconds, in 1-second steps. Longer costs more; the generate button shows the exact credits for the current settings. There is no single best length — it depends on how much story you have. Use the template lengths as starting points:

• One continuous action ("Couple from 2 Photos", "No Photo Needed"): 5 seconds.
• A move from one photo to another ("First & Last Frame"): 7 seconds.
• A story with several stages (both "Long Video" templates): 15 seconds.

Pick the time the story actually needs, starting short; if a step is skipped or flashes by in the result, give it more time. A longer video also costs more each time a try fails.

One shot or several?

If you don't say, the model decides for itself whether to cut. To keep it in one continuous take, write one continuous shot, no cuts. To control the cuts yourself — how many, and when — write the video as a shot list with timestamps: an overall line first, then one line per shot with its time range and what happens in it. Every timestamp becomes a cut.

Overall: a photorealistic intimate scene, the same two people and the same room throughout.
Shot 1 [0–3s]: side view, medium shot — the two of them on the bed, skin to skin, starting slowly.
Shot 2 [3–10s]: steady side close-up on the contact, faster and deeper; Shot 2 continues the physical rhythm established in Shot 1.
Constraint: photorealistic throughout; clothing must never come back.

Two habits that keep a shot list clean: say in each shot that it continues what the previous one started, and give the first shot the job of establishing who is there before the action begins. If a single-shot opening dissolves into mush, try splitting it into two shots this way.

8. When something goes wrong

Add the "fix-it" lines only when you hit the problem — don't put them all in from the start (see 2.2).