Seedance 2.5 — Longer, Reference-Heavy Creative Briefs
Up to 30 seconds per clip and up to 50 combined reference assets in the listed Topview workflow.


Which AI video model should you choose?
Seedance 2.5 and MiniMax H3 both support multimodal AI video generation with synchronized sound—but they fit different creative workflows.
Choose Seedance 2.5 when your brief needs up to 30 seconds in one generation, a larger reference set, or more detailed scene timing. Consider MiniMax H3 when you need shorter 5–15-second clips, an available 2K output workflow, or downloadable model weights for an eligible custom deployment.
Compare supported settings and original video examples before you generate. Neither model is automatically better for every shot.
Seedance 2.5 — Longer, Reference-Heavy Creative Briefs
Up to 30 seconds per clip and up to 50 combined reference assets in the listed Topview workflow.
MiniMax H3 — Short-Form Video and 2K Regeneration
Up to 15 seconds per clip, stereo audio, and a high-resolution regeneration path up to 2K.
The main trade-off is clip length and reference capacity versus H3's shorter workflow and its high-resolution regeneration option. The table distinguishes published model capabilities from settings listed for Topview.
| What You're Comparing | Seedance 2.5 | MiniMax H3 |
|---|---|---|
| Model developer | ByteDance | MiniMax |
| Main model category | Multimodal AI video model (hosted) | Omni-modal AI video model (open weights) |
| Duration available on Topview | 4–30 seconds | 5–15 seconds |
| Published upstream duration | Up to 30 seconds | 4–15 seconds |
| Output options listed on Topview | 480p / 720p / 1080p | 480p / 768p / 2K |
| How 2K output works | Not listed as a native Topview output tier | 768p base output followed by H3-Regenerate-2K, where supported |
| Text-to-video | Supported | Supported |
| Image-to-video | Supported | Supported |
| Reference images | Up to 30 | Up to 9 |
| Reference video clips | Up to 10 | Up to 3 |
| Reference audio clips | Up to 10 | Up to 3 |
| Total mixed reference files | Up to 50 | Up to 12 (max 12 combined across refs) |
| Reference-video duration | Combined duration subject to a 30-second limit | Combined duration up to 15 seconds |
| Reference-audio duration | Combined duration subject to a 30-second limit | Combined duration up to 15 seconds |
| Generated audio | Dialogue, music and effects in supported workflows | Native stereo audiovisual generation |
| Timing / creative direction | Time-coded shot and sound instructions | Multimodal instructions and reference-led direction |
| Editing workflows | Selected segment editing where available on Topview | Capabilities depend on the H3 variant and workflow |
| Access on Topview | Hosted generator and supported API access | Hosted generator and supported API access |
| Downloadable model weights | Not offered through this Topview workflow | Published, subject to model license and territory restrictions |
| Ideal starting point | Longer videos, larger reference packs, complex sequences | Shorter clips, 2K-oriented delivery, custom deployments where eligible |
MiniMax's official model documentation specifies a combined cap of 12 reference files, even though individual category caps add up to 15. Audio references cannot be used as the only H3 reference input. Topview web and API choices may vary by mode, account and model variant.
Sources: Seedance 2.5 Topview API · MiniMax H3 Topview API
Choose based on what the video must accomplish—not the model name.
Seedance 2.5 is a natural starting point for longer, reference-heavy scenes. MiniMax H3 deserves consideration for shorter clips, 2K delivery through its supported regeneration path, and licensed model customization.
Create an opening, development, and conclusion in one generation. This gives ads, explainers, and multi-beat narratives more room than a single 15-second clip.
Use up to 50 supported reference assets in the listed Topview workflow: 30 images, 10 videos, and 10 audio files, subject to file and total-duration restrictions. Guide characters, products, locations, motion and sound.
Use time-coded instructions to direct camera movement, actions, dialogue, music and sound effects. This is useful when specific beats must occur in a planned sequence.
Where the selected Topview workflow supports local editing, revise a targeted segment while aiming to preserve the surrounding content. Confirm the editing mode before relying on it for a production job.
Use the supported H3 high-resolution regeneration workflow to produce output up to 2K. This is worth testing for product details, cinematic compositions and other short-form deliverables where image detail matters.
Generate audiovisual clips with sound in the same workflow. H3 documents stereo output and multiple dialogue languages; always review speech accuracy and synchronization in the actual export.
Explore H3's published weights for custom infrastructure and integration, provided your intended use complies with its community license, territorial requirements, and hardware needs. Using H3 through Topview does not itself grant a self-hosting license.
When your idea fits a short duration and a smaller reference set, compare H3 before committing to a more elaborate production workflow. For a 15–30-second single generation, check Seedance 2.5 instead.
Six Creative Briefs. Two Model Outputs. Compare the Details That Matter.
The following are the six existing paired examples on Topview. They cover visual effects, spoken dialogue, multi-character action, dance, literary storytelling, and a reference-led game battle.
Each test keeps its original input brief and reference assets. Watch both videos, inspect the actions requested in the prompt, and judge whether each output fulfills the creative task. These displayed pairs are illustrative comparisons, not a statistically representative ranking of either model.
Text to Video · 15s · 16:9
A Norse-inspired hero floats over a fractured obsidian shrine as golden lightning, floating stone fragments, and glowing runes build toward the formation of a battle hammer. The original brief calls for a continuous orbital camera move, a radial shockwave, and a final wide hero shot.


Effects-heavy video can look impressive while missing the critical actions that make a shot coherent. This scenario tests whether each model coordinates camera movement, character performance, and effects over time.
Seedance 2.5 is the better fit when this shot has to escalate in one read and finish wide. The frames keep one bearded hero, move through lightning and flying debris, raise a hammer, and end on a wide view of that figure over the peak. The camera changes distance, so the orbit is not perfectly unbroken, but the action order stays readable. MiniMax H3 also produces lightning and a hammer, yet the framing jumps from a wide walk to a close-up, a vortex, and another walk, and the hero’s clothing changes between frames. Choose Seedance 2.5 for camera continuity and the planned wide ending. Choose MiniMax H3 only if a shorter effects montage is enough and cuts are acceptable.
A high-concept mythological action VFX cinematic spectacle shot on Sony Venice 2 with Cine Alta 8K sensor and Leica Summilux-C lenses, featuring photorealistic golden lightning arc flashes, dynamic ray-traced particle physics, floating terrestrial debris, and glowing ancient divine runes illuminated by hyper-detailed volumetric atmosphere. The continuous, seamless 360-degree whip-pan camera begins in a low-angle orbital sweep around a muscular, bearded Norse god suspended in mid-air above a fractured obsidian mountain shrine under a tempestuous crimson sky. His initial solemn and stoic composure dramatically shifts into furious rage as his eyes burst into blinding blue plasma energy, casting vibrant specular highlights across his weathered skin and braided hair. As the camera accelerates its continuous spiral rotation around the god, he lets out a guttural, thunderous roar, raising his right hand to summon violent streaks of celestial lightning that erupt from swirling storm clouds. The raw electrical arcs violently pull floating stone fragments and divine liquid metal together into a swirling vortex, instantly fusing them into a massive, glowing, runic-engraved battle hammer in his fist. Upon gripping the weapon, he detonates a massive radial gravity shockwave that instantly pulverizes the surrounding floating boulders into fine stardust particles and ripples through the thick atmosphere, leaving the ancient fractured shrine beneath him pulsing with deep blue subterranean magic. The unbroken dynamic shot completes its orbital loop to settle into a dramatic wide hero framing of the god hovering amidst the dissipating energy dust, legendary, majestic, thunderous, and escalating from tense stillness to an overwhelming explosive climax.
3 Image References · 15s · 16:9
Two stylized animated characters, HAORAN and XIAOYU, meet outside a café, exchange four specified English dialogue lines, and enter the building. Three images define the characters and environment. The brief also specifies the camera's approach and the outward-opening café door.


Dialogue video requires more than natural-looking faces. The model also needs to coordinate specific words, speaking roles, physical actions, and reference-based identities.
Both clips keep a stylized man in a blue sweater and a woman in a white top and pink skirt outside a café, and both travel from the street toward the door. Seedance 2.5 shows them meet, talk, walk to the steps, and the man reach the door. MiniMax H3 shows the same approach and a face-to-face exchange, then the woman turns toward the doorway while the man is still beside her. In the frames checked, neither clip clearly opens the door outward and lets XIAOYU enter first. Seedance 2.5 gets closer to that entrance. Use it when the door action is part of the brief. Use MiniMax H3 when a short café conversation is the goal. Play both with sound before you rely on the exact line order; these frames show the meeting and the approach, not a transcript.



Pixar-style 3D animation, style locked to <<<Image3>>> , <<<Image1>>> and <<<Image2>>> — render everything exactly in their look. HAORAN, the young man from <<<Image1>>> , and XIAOYU, the young woman from <<<Image2>>> , run into each other on the sidewalk in front of the café from <<<Image3>>> , both breaking into surprised, warm smiles. XIAOYU says: "Haoran? What a coincidence!" He laughs: "First time seeing you since graduation!" She replies: "Come on in, coffee's on me." He answers: "Let's go." They walk up the steps together; HAORAN pulls the right door leaf outward by its brass handle — hinges on the outer edge — and holds it open as XIAOYU steps inside, then follows her in as the door swings shut behind them. Soft rounded stylized character proportions, subsurface scattering on skin, large expressive animated eyes, warm golden-hour cinematic lighting, gentle squash-and-stretch motion. The camera tracks them from across the street with a slow push-in. Dialogue in English with accurate lip sync, no subtitles. Avoid photorealistic faces, live-action look, 2D cel-shaded style, any style shift mid-clip, doors opening inward, sliding doors, or extra background characters entering frame.
Text to Video · 15s · 16:9
A group of plush-felt animal characters rides in a car. The prompt assigns different actions to the Shiba Inu driver and passengers, including singing, taking a selfie, a duck photobomb, moving hats, flying objects, and reactions to the vehicle's motion. The camera should stay fixed inside the car.


Crowded scenes require a model to maintain several identities and overlapping actions at once. A strong opening frame does not guarantee that the same cast survives the full shot.
Seedance 2.5 is the steadier crowded shot. The interior camera stays inside the car, and the shiba driver, watermelon-hat cat, duck, raccoon, and basket animals remain recognizable while a hat comes off in the middle of the clip. Not every assigned gag stays isolated, but the cast does not get replaced. MiniMax H3 keeps the car interior and several of the same animals, yet the duck, hats, and passengers swap and new animals appear, so identity and prop continuity are weaker. Choose Seedance 2.5 when the same group has to survive the full 15 seconds. Choose MiniMax H3 only if a lively car interior matters more than each animal keeping one job.
Hyper-realistic wool-felt plush texture, cottagecore pastel palette, matching the reference image exactly. Fixed camera locked at this exact interior angle, no cuts, no camera movement. A hilarious and heartwarming 15-second road trip scene bursting with individual personality. [0-4s] The shiba inu driver is having the time of her life — head bobbing to music, singing loudly with mouth wide open, one paw drumming the steering wheel. The watermelon-hat cat films a selfie video, the duck photobombs by suddenly stretching its neck into frame with a loud honk, the cat rolls its eyes dramatically. The raccoon shakes its iced drink like a cocktail, ice cubes rattling. [4-8s] Sharp left turn! A comedic chain reaction: the croissant hat slides off the grey cat's head and lands on the bonnet kitten's face, covering its eyes — the kitten flails its tiny paws in confusion. The poodle slides across the back seat and squishes the bichon against the window, both bursting into laughter. The duck slides sideways with a deadpan expression, completely unbothered, like it has seen everything. [8-12s] Bumpy road! Everything bounces — a grape pops out of the picnic basket, flies through the air, and the raccoon catches it in its mouth mid-bounce, looking extremely proud of itself. One kitten in the pink basket gets bounced awake, grumpily grabs the edge of the basket, gives everyone an annoyed squint, then dramatically flops back down into the basket. The goldfish leaps out of the sloshing water and splashes back in. [12-15s] The road smooths out into warm golden light. All animals perk up, bright-eyed and alert, turning toward the camera with big open-eyed smiles — the shiba driver glances back with a wide grin and a thumbs-up paw, eyes sparkling; the croissant cat retrieves its hat, puts it on backwards without noticing, and blinks curiously at the camera; the bonnet kitten hiccups a tiny flower petal and giggles with wide shining eyes. Everyone stays awake, lively and energetic, eyes wide open, looking around cheerfully as sunset light glows on their fur. Every animal has exaggerated theatrical acting with natural fur physics — wide eyes, flattened ears, wagging tails, puffed cheeks, dramatic reactions, while rendering stays fully realistic plush-felt style throughout. All characters keep their eyes open and alert throughout, except the basket kittens. Blurred mountain scenery streams past the windows, tree shadows flicker across the interior.
1 Image + Audio Reference · 15s · 9:16
Five dancers follow an uploaded music track while retaining the faces, hairstyles, and outfits in a group reference image. The choreography begins in a V-formation and requires a change to a horizontal line in one continuous take.


Dance clips test simultaneous control of identity, choreography, position, and musical timing. A compelling still frame cannot prove that the motion follows the beat.
Neither clip is a reliable five-dancer lock. Seedance 2.5 starts with a small group in front of the graffiti wall and later opens into a line, but it ends on a two-person close-up, so the five-dancer requirement does not hold. MiniMax H3 begins with extra people and later swaps in a different line of dancers, so faces and outfits do not stay tied to one reference. Seedance 2.5 is the better start if you need the early group to stay smaller. Do not treat either result as beat-synced until you play it against the reference track; the frames show pose changes, not whether steps land on the beat.

Audio beat reference
The five dancers from <<<Image1>>> , keeping each dancer's exact face, hairstyle and outfit, perform street dance in the graffiti plaza, choreographed to the music in <<<Audio1>>> , every move landing on the beat. Formation change is required: starting in the V-formation from <<<Image1>>> , midway through the clip all five dancers travel to new positions — the center dancer moves to the back as the other four sweep forward into one horizontal line, finishing the routine in this new formation. Camera movement mirrors the music's rhythm and energy. Continuous single take. Use the ORIGINAL audio from <<<Audio1>>> unchanged; audio reference only, not visual reference. Avoid generated music, added sound effects, cuts, face distortion, identity swapping, subtitles, dancers standing in place for the whole clip.
Text to Video · 15s · 16:9
A watercolor-style animated adaptation of the river-boat scene from Kenneth Grahame's The Wind in the Willows. A water rat rows a blue-and-white boat while a mole sits at the stern; the boat strikes the bank and the rat tumbles backward before recovering.


Literary adaptation tests how accurately a model transforms a written event into visual action. Attractive illustrations are not a substitute for a story that actually happens as described.
Seedance 2.5 follows the passage more closely. The frames keep a watercolor river, a blue boat, a rat rowing, and a mole at the stern, then show the rat on his back in the boat before he sits up. That is rowing, a fall, and a recovery, in one painted style. MiniMax H3 also stays in watercolor with a blue boat, but the rower is dressed as a person in a striped jacket and hat, and the fall is less clearly caused by the boat hitting the bank. Choose Seedance 2.5 when character roles and story order matter. MiniMax H3 is the weaker adaptation of this passage.
A fifteen-second continuous shot, Edwardian pastoral watercolor style, soft riverside light. A water rat sculls a small rowing boat, painted blue outside and white within, along a gentle river, with a mole seated in the stern, exactly as the original text describes: "Believe me, my young friend, there is nothing—absolute nothing—half so much worth doing as simply messing about in boats. Simply messing," he went on dreamily: "messing—about—in—boats; messing—" "Look ahead, Rat!" cried the Mole suddenly. It was too late. The boat struck the bank full tilt. The dreamer, the joyous oarsman, lay on his back at the bottom of the boat, his heels in the air. "—about in boats—or with boats," the Rat went on composedly, picking himself up with a pleasant laugh." — Kenneth Grahame, The Wind in the Willows (1908), Chapter I, Project Gutenberg #27805 (Scribner 1913 ed.)
2 Image References · 15s · 16:9
TOPVIEW BOT, a small agile heroine, fights a giant toy mech in a colorful game world. The two reference images define the characters. The brief requires a connected sequence of attacks, projectiles, climbing movement, and a final hit while preserving the character designs and relative scale.


A complex battle tests character consistency, action dependencies, and spatial reasoning. A spectacular explosion is not enough if the underlying movement becomes impossible to follow.
Both clips keep a small yellow heroine against a much larger blue-and-yellow mech, so the scale contrast is readable in each. Seedance 2.5 stays on that pair through close combat, an explosion, and a wide final hit, with fewer identity changes. MiniMax H3 also shows the cannon, a climb onto the mech, and a hit, but an early frame introduces a different yellow character before the heroine returns, so the reference design is less stable. Choose Seedance 2.5 when the supplied character has to survive the whole fight. Choose MiniMax H3 if the scale contrast is the priority and you will check that the heroine still matches your reference.


Toy-world giant boss battle CG, 15 seconds, one continuous take. Input references: <<<Image1>>> for character appearance — TOPVIEW BOT PRO exact costume and palette <<<Image2>>> for boss appearance — TITAN UNIT-01 final giant form, exact design and palette References are design sheets only: do NOT render the sheet layout, labels, swatches, scale panel or grey background. Extract character design only. Character lock (match <<<Image1>>> exactly, no changes): A small agile girl TOPVIEW BOT PRO with long silver-white hair in loose low twin tails, pale mauve eyes, calm focused expression. Yellow cat-ear cap with black inner ears and black cat face emblem, black and hot pink headphones with heart marks, heart earrings. Oversized yellow bomber jacket with black cat patches, PRO badge, X badges and pink hearts, white-black cropped top, pink belt with heart buckle, black shorts, pink heart mark on her thigh, mismatched socks, chunky yellow platform sneakers with pink soles. She carries a huge yellow and black toy cannon with pink handle and X markings, connected by pink cables to a back tank. Color palette locked: yellow #FFD400, hot pink #FF4D79, black #1A1A1A, white, grey. Boss lock (match <<<Image2>>> exactly): A hundred-meter toy titan mech in its tall heroic giant form, glossy injection-molded plastic and diecast materials, dense overlapping armor plates with oversized screws and molded seams. Deep grey and matte blue armor, blue helmet with hot pink crest raised above broad flat shoulder decks, glowing yellow eye visor, shoulder missile pods with yellow-black warning stripes and pink missile tips, pink-yellow energy lines, glowing chest core with pink X emblem set into layered chest armor plates forming flat ledges at varying heights — standable plates, deep seams, no man-made stairs or ladders. The titan is an active aggressive fighter: its head tracks her constantly, its fists strike and grab at her throughout the fight, servos whining with every motion. Environment: giant toy city battlefield — building blocks, gift boxes, toy trains, a Ferris wheel, teddy bears, plastic debris; realistic grey-white skyscrapers behind, bright blue sky, floating confetti. Timeline: 0.0–3.0s | Face-off Opening frame: seen from a low angle behind her — TOPVIEW BOT PRO stands with her back to the camera on a wide flat armor ledge beside the glowing X core, tiny against the titan's mass, facing up toward the helmet far above. Wind blasts through her twin tails and yellow jacket. She levels her cannon at the distant glowing visor. The titan's head tilts down, visor locking onto her and narrowing to a bright slit; both shoulder missile pods snap open with mechanical clicks. Camera: low-angle over-the-shoulder push-in past her toward the titan's head — her back to camera in the near foreground, the helmet towering above, shallow depth of field. 3.0–8.0s | Homing missile gauntlet The shoulder pods fire: missiles arc high into the sky trailing white smoke, then curve sharply downward and home in on her, chasing her path. She sprints up the titan's chest across overlapping armor plates, vaulting ledge to ledge as the homing missiles slam into the armor just behind her footsteps, each impact exploding into colorful plastic debris. The titan's left fist sweeps across its own chest to swat her — she slides under the passing fist without breaking stride, then fires once on the run, destroying the last missile behind her. Camera: fast side-tracking shot rising with her, wide lens, heavy motion blur, the sweeping fist and missile impacts crossing the foreground. 8.0–15.0s | Face-to-face point-blank finish As she reaches the shoulder deck, the titan's right hand rushes in to grab her — she leaps off the closing fingers, using the hand as a springboard, arcs through the air IN FRONT of the titan's face, and lands on the narrow brow ledge directly above the glowing yellow visor. She faces the titan's face head-on — her body turned OPPOSITE to the direction the titan faces, her back to the open sky, the huge visor glowing right in front of her. She presses the cannon muzzle horizontally against the center of the visor. The weapon charges yellow-pink energy for one beat while the titan thrashes — then she fires the beam horizontally straight INTO the face. The visor shatters inward, colorful energy bursts through every helmet seam, and the titan breaks apart into thousands of flying toy blocks and plastic pieces. She kicks off the collapsing face through the falling debris, landing on a giant tumbling block, still running forward. Camera: following her leap around to the FRONT of the titan's face, holding a frontal close shot — girl and giant face in the same frame, facing each other — during the charge and firing, then pulling wide as the mech shatters; follow her momentum through the debris. End frame: cut during her run across the falling blocks — no victory pose, no freeze frame, no fade. Style: AAA game boss battle trailer, stylized anime-inspired 3D CG, glossy toy materials, Ratchet & Clank + Astro Bot energy, extreme scale contrast between the small girl and the colossal mech, high saturation, cinematic rim light, physical debris simulation, motion blur, 24fps anamorphic widescreen. Audio: fast electronic rock, heavy drums, toy-box chimes, servo sounds, missile launch whooshes, homing missile whistles, sliding metal scrapes, cannon charge whine, massive beam impact, cascading plastic destruction; music cuts off mid-action. Avoid static camera, slow pacing, scene reset, character facing the camera in the opening shot, girl facing the same direction as the titan's head, girl standing on top of the head firing outward into empty sky, beam firing away from the titan, missiles flying away in straight lines, missiles appearing from off-screen without being launched, passive motionless boss, staircases, man-made steps, ladders, handrails, character redesign, hair or costume color change, blue jacket, bunny ears, weapon redesign, chibi proportions on the mech, rendering the reference sheet layout or text, extra limbs, distorted face, bad hands, subtitles, watermark, UI, gore, desaturated colors, slow-motion ending.
We compare the task the creator requested with the result each model produced.
Each scenario has its own original prompt and, where applicable, original image or audio references. The same intended brief should be used for both models. Model-specific input labels may differ; these differences had been recorded.
Record the actual duration, aspect ratio, resolution, model ID, generation mode, and audio setting for each output. Equal requested duration does not guarantee equally configured exports, especially when the models offer different resolution tiers.
Review whether the specified events, actors, objects, camera movements, and endings appear in the correct order.
For reference-led cases, check identifiable faces, outfits, products, props, and environment details against the source assets.
Evaluate camera changes, body motion, object trajectories, spatial relationships, and whether the scene resets unexpectedly.
Where audio is part of the task, review spoken-word accuracy, speaking roles, lip synchronization, music continuity, and synchronization with action.
Distinguish a technically completed render from a video that meets the intended creative requirements. Failed tasks, incomplete outputs, and quality rejections should be tracked separately.
The existing clips are displayed as paired examples. We set the number of generations at 52 in total and selection rule in advance and retain all attempts. The displayed video is the median-scoring output.
Use the same intended task, then compare the credit quote and usable result.
| Published Output Tier | Seedance 2.5 on Topview | MiniMax H3 on Topview |
|---|---|---|
| 480p | 0.7 credits/sec | 0.15 credits/sec |
| 720p (Seedance) / 768p (H3) | 1.5 credits/sec | 0.3 credits/sec |
| 1080p (Seedance) / 2K (H3) | 3.7 credits/sec | 0.6 credits/sec |
The middle and upper tiers are not identical resolutions, and H3's 2K delivery involves a regeneration process. These examples compare published charges, not image quality at the same pixel count or production cost across identical technical conditions.
The lowest price per generation is not necessarily the lowest price per finished project. Track every paid attempt and how many results meet your quality requirements.
Cost per usable output = Total credits consumed ÷ Number of usable videos.
A fair cost comparison also needs the workflow, account plan, task duration, output settings, and number of attempts.
MiniMax H3 is worth testing when your output target matters more than a longer clip or a large reference pack.
If your shot fits within 15 seconds and you want to evaluate the 2K regeneration workflow, compare H3's output against the detail you actually need in the final delivery.
Downloadable weights may appeal to technical teams building internal systems. Check the current community license, excluded territories, additional permissions, and hardware requirements before deploying or distributing outputs.
H3 can combine supported text, image, video, and audio context within its limits. A smaller, well-structured reference set may be enough for a short character scene, product concept, or audiovisual experiment.
You need one continuous 16–30-second generation, more reference assets than H3 supports, or a selected-segment editing workflow available through Topview.
One workspace. Two different ways to create.
Choose Seedance 2.5 for longer scenes and reference-heavy production. Explore MiniMax H3 for shorter audiovisual work, an available 2K output path, or eligible open-weight integrations.
Review the settings, check the credits, and compare the results for your actual creative task.