Portrait and character close-up work is where AI video prompts fail most visibly. The problem is almost never a lack of descriptive language — it is a lack of constraint architecture. A prompt that writes "close-up of a woman, emotional, cinematic lighting" gives Seedance no information about which emotion, which lighting direction, which camera behavior, or how to connect one shot to the next. The model produces something plausible and generic. A prompt that writes "[0–3s] Camera: slow imperceptible push-in. Action: eyes crinkle at the corners and cheeks lift. Lighting: warm golden-hour glow, frontal" gives Seedance a complete instruction set for a single beat — and that specificity is what makes the result look directed rather than generated.
The five prompts below cover five entirely different portrait strategies. The 360-degree orbit solves character consistency by treating the subject like a product. The emotion-color mapping table solves multi-shot lighting by replacing per-shot descriptions with a four-line rule. The GRWM vlog-cinematic hybrid stress-tests identity coherence across 10–15 shots and multiple environments. The character sheet reveal structure teaches the detail→identity→presence→full-reveal grammar for introducing someone the audience hasn't met. And the five-phase emotional close-up shows what it takes to get a 15-second face-only shot to sustain five distinct physiological states without identity drift.
Here are 5 Seedance portrait and character close-up prompts from the Scenic gallery — a 360-degree orbit for all-angles character identity, emotion-color logic for music-video-style portraits, the GRWM multi-shot consistency challenge, the classic character sheet reveal structure, and a five-phase emotional close-up grid. All free to copy.
1. The 360-degree character orbit — product photography logic applied to portrait
See the full prompt on scenic.sh →
"CAMERA: Continuous smooth 360-degree orbit around the subject at constant speed and constant chest height. Starting from the front, moving clockwise through left, back, right, and returning to front. The orbit takes the full 15 seconds."
Why this works: At 64 likes — the highest-engagement prompt in this set — this works because it borrows the grammar of product photography and applies it to a character. Product shots use orbits to expose every angle of an object without opinion; this prompt uses the exact same logic on a human subject. "STYLE: Photorealistic editorial. Shot on 35mm, f2.8. Neutral grey studio backdrop. Even, clean lighting that reveals form from every angle. No atmospheric effects. Deep DOF." removes all cinematic mood cues — no bokeh, no drama, no lighting variation — forcing the model to make the character interesting through form alone. "He stands completely still in a neutral pose, centered in the frame. No movement whatsoever. Arms relaxed at sides." is the performance instruction: the subject is not performing, not reacting, not emoting. The camera is the active element; the character is the stable reference. "Constant speed and constant chest height" makes the orbit a measurement instrument rather than a cinematic move — the viewer experiences all angles at the same distance and elevation, making the character's geometry, silhouette, and costume readable from every direction.
The takeaway: use a product-shot orbit with no atmospheric effects and deep DOF to create an all-angles character identity reference. "Constant speed, constant chest height, starting from front, moving clockwise" is a fully specified mechanical instruction that removes camera variability — every frame of the orbit contributes equally to character recognition. This is the correct prompt structure for building a consistent character reference you can use across multiple scenes or future videos.
2. The emotion-color mapping portrait — a four-line table that replaces per-shot lighting
See the full prompt on scenic.sh →
"COLOR LOGIC: Calm = blue / Anger = red / Joy = gold / Sad = desaturated / STYLE: Cinematic emotional realism / SHOT FLOW: Neutral state (soft tone) → Phone notification → color shifts red → Frustration builds → Deep breath → fades to blue → Smile moment → warm gold → Sudden memory → muted gray → Close-up eyes → Rapid color shifts → Stabilization → Final calm expression"
Why this works: At 53 likes, this prompt's central innovation is the COLOR LOGIC table — four lines that replace individual lighting specifications for all 11 shots. Instead of writing "warm amber lighting for the joy beat" and "cold blue sidelighting for the calm beat" and "harsh red fill for the anger beat," you write a four-line rule and let Seedance apply it globally. "FORMAT: 15s / 120 BPM / 11 SHOTS" tells the model the edit tempo: 11 shots in 15 seconds at 120 BPM means roughly one shot per beat, creating a music-video rhythm rather than a cinematic one. "ENVIRONMENT: Same location, color changes based on emotion" is the efficiency move — a single set re-lit through the COLOR LOGIC rule rather than 11 separate scenes, so the emotional progression reads as continuous. The narrative trigger "Phone notification → color shifts red" grounds the first abstract color shift in a concrete real-world event, giving the audience a legible cause before the metaphor takes over. "Close-up eyes" midway through the shot list is a beat marker that resets the emotional register before the rapid color shifts — it's a rest before the finale.
The takeaway: build a color-logic table (four entries: emotion = color temperature) instead of per-shot lighting descriptions. This is a global rule that Seedance applies consistently across all shots, producing a coherent visual language without the risk of per-shot descriptions contradicting each other. The rule should include at least one narrative anchor — a concrete real-world trigger for the first color shift — so the opening emotion change is motivated, not arbitrary.
3. The GRWM vlog-cinematic hybrid — multi-shot character consistency across environments
See the full prompt on scenic.sh →
"A young woman matching the exact facial features, hairstyle, and identity of the reference image is filming a 'Get Ready With Me' style video in a softly lit, modern bedroom. Cinematic but vlog-inspired, smooth camera motion, shallow depth of field. Maintain facial consistency across all shots."
Why this works: At 51 likes, this prompt works because it deliberately creates the hardest portrait consistency challenge — 10–15 shots across three different environments (bedroom, sink, vanity) with different lighting conditions, different clothing states, and a mid-video wardrobe change — and then asks Seedance to hold the same character identity throughout. The shot list reads as a temporal arc: "Close-up of her tying her hair into a loose bun / Applying face wash at the sink, soft water splashes / Gently patting her face dry with a towel / Skincare routine in front of mirror / Sitting at vanity applying makeup step-by-step." Each action is contingent on the previous one — the hair goes up before the face wash, the face is dried before skincare, skincare happens before makeup — creating a continuity chain that forces Seedance to track the character's state across scenes rather than treating each shot independently. The "Final reveal scene" (stepping back in a red outfit after a full transformation) works because the reveal's impact depends on everything that came before. "Cinematic but vlog-inspired" is a hybrid style instruction that sits between two extremes: the vlog register (casual, reactive, direct-to-camera) provides the energy; the cinematic register (shallow depth of field, smooth motion, soft highlights) provides the quality.
The takeaway: structure a multi-shot portrait as a temporal arc with contingent actions — each shot depends on the state established by the previous one, forcing the model to maintain character continuity across different environments and lighting conditions. The GRWM format is the optimal container for this because the actions form a natural causal chain. Include "Maintain facial consistency across all shots" as an explicit instruction, not an assumption — it tells Seedance that identity coherence is the primary goal, more important than cinematic variety.
4. The character sheet reveal — detail → identity → presence → full reveal
See the full prompt on scenic.sh →
"Structure: detail → identity → presence → full reveal. Make the character active: she moves, reacts, interacts with her environment and prop while talking — short, natural gestures, small shifts, purposeful motion. Show acting range: subtle emotional shifts while speaking (confidence, hesitation, curiosity, intensity)."
Why this works: At 12 likes, this prompt teaches the classic cinematic grammar for character introduction — the same structure used in film to introduce a character the audience hasn't met. Starting with details (outfit, prop, hands) before revealing the face forces Seedance to build character through specificity rather than frontal presentation; by the time the audience sees the whole frame, they've already registered who this person is through fragments. "Open with the character looking into camera and speaking naturally, introducing herself in her own words" immediately establishes direct address — she is aware of the camera, making this feel like an audition or interview rather than observation. The critical instruction is "Do not treat the sheet as a single image. Use its elements as separate shots." This converts a static reference image into a shot list: each element of the sheet becomes its own framing source. "Show acting range: subtle emotional shifts while speaking (confidence, hesitation, curiosity, intensity)" names four distinct emotional registers in sequence — the model is directed to move through them rather than hold a single expression. "Controlled, minimal movement (soft push-ins, light tracking, subtle handheld)" gives the camera behavior: no aggressive moves that would compete with the performance; the camera tracks, it doesn't direct.
The takeaway: the detail → identity → presence → full reveal arc is a character introduction grammar — it forces Seedance to build the character from the outside in, establishing physicality before showing the complete person. Instructing the model to treat the reference sheet as separate shot sources rather than a single image is the parsing move that unlocks the structure. The four-register acting direction (confidence → hesitation → curiosity → intensity) is a progression, not a mood board — name them in order so the model understands each beat transitions into the next.
5. The five-phase emotional close-up — triple specification per beat for face-only work
See the full prompt on scenic.sh →
"[0-3s] Camera: Slow, imperceptible push-in. Action: The face breaks into a genuine, soft smile; eyes crinkle at the corners and the cheeks lift. Lighting: Warm golden-hour glow, soft and frontal. [3-6s] Camera: Static extreme close-up. Action: The smile dissolves into a heavy, downward curve; eyes well up with glistening tears. Lighting: Transition to a cool, melancholy blue wash from above."
Why this works: Despite 2 likes, this is the most technically rigorous prompt in the set for pure face work. Each of the five phases has three independent specifications: a camera instruction, an action, and a lighting condition. "[0-3s] Camera: Slow, imperceptible push-in. Action: The face breaks into a genuine, soft smile; eyes crinkle at the corners and the cheeks lift. Lighting: Warm golden-hour glow, soft and frontal." — this triple specification gives Seedance no ambiguity about what the model is doing (pushing in), what the face is doing (smiling with specific muscular detail), and what the light is doing (warm, frontal). The five-phase emotional arc — joy → grief → anger → fear → serenity — is paired with a corresponding lighting progression: warm gold → cool blue → red/orange sidelight → dim flickering → diffused white. The lighting temperatures encode the emotional states, so even a viewer watching without the action track would read the emotional arc from the color temperature alone. "Avoid: Identity drift, jitter, distorted limbs, unnatural morphing artifacts" is a targeted negative constraint list for a 15-second face-only close-up — these are the four specific failure modes of this shot type, not generic quality requests.
The takeaway: for a 15-second face-only close-up, specify camera + action + lighting as three independent parameters for each beat — never combine them into a single sentence. The triple-specification approach removes the ambiguity that causes generic results: the model knows what the camera is doing, what the face is doing, and what the light is doing at each moment. Build a per-phase lighting temperature arc (warm gold → cool blue → red sidelight → dim flicker → diffused white) that mirrors the emotional arc, so the light narrates the progression independently of the action track.
Portrait & character close-up prompt cheat sheet
Across all five, the structural techniques that make Seedance portrait and character close-up prompts work:
- Use a product-shot orbit for character identity — "constant speed, constant chest height, starting from front, clockwise, 15 seconds" treats the character the same way a product shoot treats an object: all angles equally, no cinematic bias. Remove atmospheric effects and deep DOF forces form to carry the identity.
- Replace per-shot lighting with a color-logic table — four lines (Calm = blue, Anger = red, Joy = gold, Sad = desaturated) let Seedance apply a consistent lighting language across all shots without per-shot specification. Include one narrative trigger to anchor the first color shift in a real-world event.
- Structure multi-shot portraits as temporal arcs with contingent actions — each shot depends on the state established by the previous one. The GRWM sequence forces character identity to hold across environments and lighting conditions because the actions are causally linked. State "Maintain facial consistency across all shots" explicitly.
- Use detail → identity → presence → full reveal as the character introduction grammar — start with fragments (hands, outfit, prop) before the face; instruct the model to treat reference sheets as individual shot sources; name emotional registers in progression order.
- Triple-specify each beat for face-only close-ups — camera instruction + physiological action + lighting condition as three independent parameters. Pair the emotional arc with a corresponding lighting temperature arc. Target negative constraints at the specific failure modes of the shot type (identity drift, jitter).
Browse the full Scenic portrait and character gallery or read how to write Seedance 2 prompts for the complete cinematic prompting guide.
