Higgsfield 新手教程：2026 年如何制作 AI 视频

To make your first AI video, upload a photo or write a short prompt, pick a model, set the length and resolution, and press Generate. On Higgsfield, you do all of this in Video → Create Video, and the example below walks through one real 10-second clip from source image to download. Many weak outputs trace back to a decision made before hitting generate: a vague prompt, a blurry source image, the wrong model for the job, or generating at full resolution before confirming the prompt works. The model works with what you give it.

Which AI Video Tool Should You Start With?

Pick the model that fits the shot you need. The table below covers common beginner goals, the model to try first, and the Higgsfield plan where it’s available.

Model availability checked on October 29, 2026. Plans and model access may change.

What Is AI Video Generation and How Does It Work?

AI video generation takes a text prompt, an image, or both, and produces a short video clip. You describe what you want, the subject, action, setting, lighting, and camera angle, and the model generates it frame by frame.

There are two workflows. Text-to-video means you write a prompt and get a video. Image-to-video means you upload a photo and the model animates it. Many beginners start with image-to-video because the output is often more predictable: the model starts from a real picture instead of interpreting everything from text alone.

The model generates the whole clip in one pass, so a single clip usually holds together better than two separate clips. Still, faces, hands, and objects can drift or change mid-clip. The bigger challenge starts when you need a second clip. Each new generation starts from scratch. Most beginners run into this after their first few clips.

What Do You Need to Get Started?

You need three things: a clear idea of the shot you want, a prompt that describes it, and a source image if you are doing image-to-video. What goes into the prompt depends on the mode.

Text-to-video: describe the scene and the motion. The model has nothing to look at, so the prompt has to cover everything:

• Subject: who or what is in the frame
• Action: what they are doing
• Setting: where it is happening
• Camera: how the shot is framed and how it moves

Weak: “A woman walking.” Stronger: “A woman in a red coat walks through a rainy Tokyo street at night, close-up, slow motion, neon reflections on wet pavement.” The more specific you are, the less the model guesses.

Image-to-video: describe the motion first. The image already sets the subject, setting, and look, so the prompt mostly needs to say what moves, how it moves, and what the camera does. Repeating what’s already visible in the image adds little.

A good source image for image-to-video:

• Sharp focus and clear lighting
• The subject that matters is clearly visible; if it’s a person, the face isn’t hidden
• No heavy clutter obscuring what matters
• Resolution at least as high as the video you plan to make; check the model’s upload limits

Side angles, dark lighting, and busy backgrounds often make results less predictable. You can still use them, but simpler shots are easier to start with. Don’t have a photo that fits? You can generate a source image instead, for example in Soul Cinema on Higgsfield.

How to Generate Your First AI Video: Step by Step

Below are the six steps, with one real example made on Higgsfield: a still frame created in Soul Cinema turned into a 10-second, three-shot clip with Seedance 2.5.

Step 1. Pick the right workflow. Image-to-video for realistic people, product demos, or any shot where you want a specific face or object. Text-to-video for abstract scenes, landscapes, motion experiments, or anything where you do not have a reference image. Open Video → Create Video and upload the image.

In our example: the reference image is a still frame we generated in Soul Cinema: a rider and a passenger on a red motorcycle with a sidecar under an elevated train line.

Step 2. Write your prompt like a director. For text-to-video, describe the subject, action, setting, lighting, and camera in one or two sentences. For image-to-video, focus on the motion of the subject and the camera. Add a mood or style if it matters: cinematic, documentary, dreamlike. Avoid adjectives like “beautiful” or “amazing”; they give the model nothing to work with.

In our example: Our full prompt has three shots, the camera path, the light, and the sound second by second. For your first clip, two sentences about the motion are enough, for example: “The rider pulls the goggles down and starts the engine. The camera slowly pushes in as the passenger smiles at him.”

Reference image: @image1 KEYFRAME — the single anchor for the passenger's face, hair and wardrobe, the rider's wardrobe, helmet and goggles, the motorcycle with sidecar, the street, the light and the color grade. @image1 is the LAST moment of Shot 1 (≈4.8 s), just before the hard cut to the kick-start insert: the motorcycle parked at the curb facing screen-right, side-on to camera, sidecar toward camera; the passenger settled in the sidecar at frame left with one arm resting on the rim, glancing at the rider with a small smile; the rider seated on the saddle, cream helmet on, chrome goggles over his eyes, both hands on the grips. The video does NOT start on this frame — Shot 1 opens low at the foot of the stairs and the camera travels until it lands on exactly this composition. In @image1 the rider's face and hair are hidden by the helmet and goggles — they are described in text below. The vendor, the train, the traffic and the kick-start are not in @image1 and are described in text. **Format:** 10 seconds, 16:9, 24fps, three shots. **Style:** 8K IMAX. Photorealistic — no 3D render, no game engine, no game-cutscene aesthetic. Premium urban motorcycle commercial with a real practical riding feel and a loose, playful energy — a sidecar motorcycle weaving through dense traffic on a sunlit avenue under a vintage elevated train line in a classic East Coast American city, with the warm texture of classic American street films. **Cinematography:** Claudio Miranda kinetic vehicle cinematography with Emmanuel Lubezki's long-take precision. Three shots: a continuous move from the stairs to the saddle that lands on @image1, a punchy close insert of the kick-start, then one continuous, highly active chase move that swings around the moving motorcycle — in front of it, around its sidecar side, behind and above at car-roof height, and back in front — always within arm's reach, always revealing the zig-zag line through the traffic. **Lighting:** Natural daylight only, matched to @image1. Soft open shade under the tracks at the stairs and the curb; out on the avenue warm afternoon sun from the right, about 4500K, cut by the elevated tracks into sharp diagonal stripes of light and shadow that slide over the rider, the passenger, the motorcycle and the cars as they move — from above they read like a moving barcode; manhole steam glowing where the sun hits it; warm rim light catching the passenger's platinum hair; specular glints on chrome, car roofs and flanks, the stainless-steel cart and the goggle lenses. Subtle anamorphic flare when the lens swings past the sun between the girders. No artificial lights. **Color:** 60:30:10 — 60% warm city neutrals: cream and buff facades, grey asphalt, dark-green riveted steel / 30% stainless silver and chrome, pale sky blue, the rider's ecru and white streetwear, the passenger's platinum hair and black clothing, green awnings / 10% deep burgundy-red motorcycle and sidecar rhyming with the passenger's red racing jacket, with pops of taxi yellow and the blue-and-white umbrella. Exactly the grade of @image1. Smooth highlight roll-off, open shadows, no crushed blacks. **Camera:** One physical 24mm cine lens for shots 1 and 3, a 50mm macro for the shot 2 insert. Shot 1: low start at the foot of the stairs tilted up, then a tilt up and a backward travel just ahead of the rider at chest height past the hot dog cart, then a lateral move to the right around the parked motorcycle to its sidecar side, settling side-on at sidecar height on exactly the composition of @image1. Shot 3: low backward travel in front of the moving motorcycle → a fast arc around its sidecar side to the rear → riding behind and above at car-roof height → a rise and swing around the corner of a delivery truck looking down on the zig-zag → a drop back in front, traveling backward at handlebar height to the end. Camera distance rule: the camera always stays within a few meters of the rider and the passenger — no wide establishing views, no high aerial views, no pull-outs; both stay large in frame. 180-degree shutter motion blur. Real-time speed only — no slow motion, no speed ramps. The camera is never visible: no drones, no camera cars, no rigs, no crew, no equipment anywhere in frame or in reflections. **Skin:** Pore-level realism — vellus hair, visible pores, natural flush. The rider: small dark moles scattered on both cheeks and the faint mustache-and-chin stubble shadow preserved exactly. The passenger exactly as in @image1: fair skin with faint freckles across the nose and cheeks, small silver rhinestone face gems under both eyes, smudged smoky-brown eye makeup and matte coral lips. Pore-shadow matching the striped light. No airbrushing, no smoothing. **Acting:** Loose, playful, effortless — two friends who do this every day and are having fun, not straining. The rider: easy swagger, relaxed shoulders, a crooked grin; empty-handed on the stairs; he takes the helmet from the vendor's hand in stride with a quick nod, looks down at it with a grin as he walks, pulls it on, and once seated pulls the goggles down over his eyes; riding, he sits upright and relaxed, elbows loose, carving through traffic with small, quick, confident moves of the bars — cool, never frantic. No hand signs, no salutes, no gestures. The vendor: a warm, knowing grin, as if he does it every day. The passenger: carefree and cheeky — she follows him down the stairs, walks around to the sidecar, steps in and settles back with one arm on the rim, giving the rider one short sideways glance with a small smile, exactly as in @image1; through every cut she throws her weight into the turn like a sidecar racer, grinning. Eye-line: the rider looks at the stairs, the vendor, the helmet, the traffic and the road — never into the camera, never at the passenger; apart from that one glance in @image1 the passenger looks at the stairs, the traffic and the road — never into the camera. The vendor, drivers and pedestrians never look at the camera. **Physics:** Gravity and inertia respected — real weight on the steel stairs; the helmet has real weight in the hand-off. Mounting rule: the rider mounts like a real motorcyclist — standing on the motorcycle's left side, left hand on the left grip, he swings his right leg high over the rear of the saddle, sits down in the middle of the saddle facing forward with both feet planted on the ground — never jumping, never vaulting, never sitting sideways, never climbing over the tank; the suspension compresses naturally under him; the passenger steps into the sidecar with one hand on the rim and drops into the seat while the sidecar settles only slightly; the kick-start is one clean stamp of his right foot. Riding rule: the motorcycle never rides in a straight line for more than a moment — it weaves through the dense, slow two-lane traffic in a continuous zig-zag, clearly changing lanes again and again, with clear space each time, no contact, nobody forced to brake hard; it stays on the avenue under the elevated tracks; the sidecar rig swerves flat and never leans — the passenger shifts her own weight into each cut instead; the tires bite and scrub lightly; the cars move steadily with real mass; both jackets flutter; the passenger's platinum hair whips in the wind; steam tears apart in the motorcycle's wake. Correct contact shadows for sneakers, boots and tires. No floating props, no sliding feet. **Composition:** The rider is the anchor of every frame — centered or on a third, in focus. In shot 3 the cars are always close around the motorcycle: car flanks, bumpers, roofs and riveted columns wipe through the foreground and frame it, so every change of lane reads instantly; from above the zig-zag line between the cars is unmistakable, with the stripes of shadow sliding over everything. Passenger placement: the camera favors the sidecar side and the front so the passenger is clearly visible beside the rider, never overlapping him — she never covers his face or upper body; on the stairs and the sidewalk she stays half a step behind and to his side. The rider always holds the focus. **Continuity:** Rider, passenger, motorcycle, cars and street identical in every cut and exactly as in @image1. The rider: same face, moles and short crop until the helmet goes on; same ecru jacket, white tank top, khaki-grey cargo pants, white trainers; empty-handed until he takes the helmet from the vendor in shot 1, helmet on from then on; the chrome goggles are strapped around the helmet and he pulls them down over his eyes once seated. The helmet is in the vendor's hand at the cart until the hand-off. The passenger: same face, face gems and platinum wolf cut with dark roots, same red-and-black racing jacket, black top, black choker, black cargo pants and black combat boots, bare-headed throughout; she rides in the sidecar from the end of shot 1 to the last frame. The sidecar is always on the motorcycle's right side. The same silver train on the tracks overhead. Every window in the street is plain glass with curtains or dark interiors — nothing displayed in any window. The rider and the passenger never touch or speak to each other; the passenger touches only the stair rail, the sidecar's rim and grab rail. No identity drift, no morphing. **Editing:** Three shots, real time. Shot 1 (0.0–4.9 s) — one continuous move from the stairs to the saddle, ending on @image1. Hard cut to shot 2 (4.9–5.3 s) — a close insert of the kick-start. Hard cut on the engine bark to shot 3 (5.3–10.0 s) — one continuous, highly active chase move to the end. No dissolves, no slow motion. **Technical:** 24fps, 180-degree shutter, real-time playback only, light 35mm film grain, subtle anamorphic flare. 8K detail. No jitter, no warping, no morphing, no frame blending. No drones or camera equipment visible. **Audio:** Environmental SFX only. No music. The train's steel rumble overhead, sneakers and boots ringing on diamond-plate steps, the hot dog cart sizzling, the soft slap of the helmet into the rider's hand, helmet strap click, goggle-strap snap, kick-start lever clank, a twin-cylinder engine barking to life with exhaust pops echoing under the tracks, the engine note rising and dipping with every cut through the traffic, light tire scrub, cars whooshing past inches away, one short generic taxi horn, wind, soft unintelligible street chatter. No voices, no intelligible dialogue, no subtitles. **Character — Rider (helmet, goggles and wardrobe exactly as in @image1; his face is described here because @image1 hides it):** East Asian man in his mid-twenties, light warm skin with a natural cheek flush, a few small dark moles scattered on both cheeks, faint mustache-and-chin stubble shadow; lean face with a defined jawline; dark brown almond eyes; straight dark brows; thick dark-brown textured short crop with a choppy short fringe above the brows and tapered sides; small silver hoop earrings. Lean build, five fingers per hand. Wardrobe: cropped boxy ecru-cream cotton-canvas work jacket with a corduroy collar, two chest patch pockets and a snap front, worn open; white ribbed tank top; silver chain necklace; khaki-grey wide-leg cargo pants with side cargo pockets; white leather low-top trainers with gum soles, no logos. Helmet: glossy cream open-face helmet with brown leather lining and a cream chin strap. Goggles: vintage aviator riding goggles with polished chrome frames and tinted lenses, strapped around the helmet. **Character — Passenger (exactly as in @image1):** East Asian woman in her early twenties, fair skin with faint freckles; soft oval face; brown eyes with smudged smoky warm-brown eyeliner; matte coral-pink lips; small silver rhinestone face gems under both eyes; small silver hoop earrings. Platinum-blonde shaggy layered wolf cut with visible dark roots and wispy bangs. Slim build, five fingers per hand. Wardrobe: oversized cropped red racing-style bomber jacket with black side and sleeve panels and white piping, worn open, no logos; black crew-neck top; black leather choker; black baggy cargo pants; black chunky lace-up combat boots. Bare-headed. Original generic character, no celebrity likeness. **Character — Hot dog vendor:** Man in his fifties with a friendly round face, short grey hair and a grey moustache, white short-sleeve shirt, white apron, dark trousers. Original generic character, no celebrity likeness. **Location (as in @image1):** A sunlit avenue under a vintage dark-green riveted steel elevated train line with lattice girders and riveted columns. The cream corner building with the produce and flower stand under the dark-green awning with the fictional wordmark seen in @image1; the dark-green steel staircase with diamond-plate treads and green steel handrails rising to the elevated station; the row of blue and green newspaper boxes; a young street tree in an iron grate; a stainless-steel hot dog cart under a blue-and-white striped umbrella at the foot of the stairs. The motorcycle waits at the curb pointing down the avenue to the right, sidecar on the street side. Across the avenue: a streamlined stainless-steel diner, storefronts with green awnings, cream, buff and brick walk-ups with fire escapes. The avenue is busy with dense, slow two-lane traffic moving in the same direction, slower than the motorcycle, with gaps opening and closing: yellow taxi sedans, a burgundy vintage sedan with chrome trim, a dark vintage sedan, a cream boxy delivery truck with plain unmarked sides; plumes of white steam rising from manholes; a few pedestrians. The train: silver stainless-steel commuter cars, plain sides, no logos, no route signs. All signage as fictional made-up wordmarks; no real brands, no logos, no license plate text. **Hero prop — Motorcycle with sidecar (exactly as in @image1):** Classic 1960s-style motorcycle in deep glossy burgundy-red: teardrop fuel tank with a cream oval knee panel, round chrome headlight in a burgundy shell, round chrome mirrors on tall stems, chrome handlebars with black grips, tan leather bench saddle, polished twin-cylinder engine, chrome kick-start lever on the right side, low chrome exhaust pipes, chrome spoked wheels, burgundy mudguards. Sidecar on the right side: burgundy torpedo-shaped tub with a thin cream pinstripe, curved clear windscreen in a chrome frame, tan leather seat and padded rim, chrome grab rail, its own chrome spoked wheel. No badges, no logos, no license plate text. **Mood & tempo:** 10 seconds, real time, always up close: an easy walk down the stairs → the vendor's helmet hand-off → a practised mount landing on @image1 → a punchy close ignition → a wild, joyful zig-zag through the cars with the camera swirling around them → a breezy finish rushing toward the lens. Fast and fun, never tense. **Shots:** Shot 1 — continuous move (0.0–4.9 s): (0.0–0.8 s) The camera starts low at the foot of the green diamond-plate stairs, tilted up: the rider's khaki-grey cargo pants and white trainers come down the steps toward the lens, hands free; behind him at left the passenger's black combat boots and red jacket follow; a silver train glides along the elevated tracks overhead. (0.8–1.6 s) The camera tilts up and begins traveling backward as the rider reaches the sidewalk: from behind the hot dog cart at right, under the blue-and-white umbrella, the vendor holds out the cream helmet with a knowing grin and the rider takes it from his hand in stride with his right hand and a quick nod; the passenger comes down the last steps behind him. (1.6–2.8 s) The camera travels backward just ahead of him at chest height: he walks toward the lens holding the helmet in both hands, looking down at it with a crooked grin; the passenger walks half a step behind at his side by the stair rail, the produce stand and its awning behind her. (2.8–3.4 s) He pulls the helmet on, the goggles strapped around it, as the parked motorcycle slides into the foreground; the camera moves laterally to the right around it toward its sidecar side; the passenger peels off toward the sidecar. (3.4–3.9 s) The rider, at the motorcycle's left side with his left hand on the left grip, swings his right leg high over the rear of the saddle and sits down facing forward, feet on the ground; at the same time the passenger steps into the sidecar with one hand on the rim. (3.9–4.9 s) The camera settles side-on at sidecar height: the rider pulls the chrome goggles down over his eyes with both hands, then puts both hands on the grips and looks ahead; the passenger drops into the seat, settles back with one arm on the rim and gives him a short sideways glance with a small smile — the frame lands exactly on @image1. Hard cut. Shot 2 — close insert (4.9–5.3 s): 50mm macro low beside the engine: his white trainer with its gum sole stamps the chrome kick-start lever down in one clean move; the polished engine shudders and barks to life. Hard cut on the bark. Shot 3 — continuous active chase (5.3–10.0 s): (5.3–5.8 s) The camera travels backward low in front of the motorcycle as it pulls away down the avenue straight toward the lens, yellow taxis on both sides, a dark vintage sedan behind, the stripes of shadow from the tracks sliding over everything. (5.8–6.4 s) The motorcycle cuts hard left past a yellow taxi; the camera whips in a fast arc around its sidecar side to the rear, the taxi's yellow flank and the sidecar sliding past inches from the lens in heavy motion blur; the passenger throws her weight into the turn, grinning. (6.4–7.5 s) The camera rides behind and above at car-roof height as the motorcycle zig-zags between a burgundy vintage sedan on the left and a yellow taxi on the right, then swings left around the slow cream delivery truck ahead. (7.5–8.3 s) The camera rises and swings around the truck's rear corner, looking down as the motorcycle carves right across an open stretch of asphalt striped with sun and shadow — a clear S-shaped line through the cars. (8.3–10.0 s) The camera drops in front and travels backward at handlebar height: the motorcycle comes straight toward the lens, the rider centered, relaxed, both hands easy on the grips, the passenger beside him in the sidecar at frame left, grinning, taxis and vintage sedans sliding past on both sides, the elevated tracks overhead. The shot ends close on the rider, still moving — no pull-out.

Step 3. Choose a model matched to your shot. Use the table above. For a first test, pick a model that supports your mode and lets you make a short, low-resolution clip.

Step 4. Start at lower resolution. Most models let you choose resolution or a quality level. Start at 720p for your first few generations. It costs fewer credits, generates faster, and tells you whether your prompt is working before you spend credits on a full-quality version. The exact price for your settings is shown on the Generate button before you start.

In our example: we generated the final clip at 1080p, and the Generate button showed 160 credits. When you’re still testing a prompt, start at 720p to spend fewer credits. For typical wait times, see How Long Does AI Video Generation Take.

Step 5. Review the full clip. Check whether the motion holds throughout. Check whether expressions or movements still feel natural near the end of the clip, not just at the start. Many quality problems appear mid-clip, not at the opening frame. If the clip works, download it from your generation history.

Step 6. Change one thing per iteration. If something is off, change the prompt, the source image, or the model, one at a time. Changing all three at once makes it impossible to know what fixed it.

In our example: in the short kick-start insert, the rider’s foot rests on the footpeg instead of stamping the kick-start lever. One change to try: describe that insert in its own short paragraph so the instruction doesn’t get lost in a long prompt. A retry with the same settings costs another 160 credits.

What Are the Most Common Beginner Mistakes?

If you’re struggling to write a detailed prompt, describe your idea in plain language to an AI assistant like Claude or ChatGPT and ask it to turn it into a video generation prompt. It gives you a first draft to edit.

What Does AI Video Generation Actually Cost?

On Higgsfield, every generation costs credits. The price depends on the model, clip length, resolution, and options like sound, and the exact amount is shown on the Generate button before you start.

Our example clip, Seedance 2.5, 10 seconds, 1080p, sound on, one reference image, cost 160 credits, and each retry with the same settings will cost the same amount. Higgsfield’s Plus Plan costs $49 per month for 1,000 credits, enough for about 6 clips like our example.

What to know before you pick a plan:

• Veo 3.1 is available on Plus and Ultra.
• On Starter, Kling 3.0 runs at 720p only; 1080p and 4K start from Plus.
• First-and-last-frame control is available from Plus.

Prices checked on October 29, 2026. Prices and plan details may change.

What Are the Limits of AI Video in 2026?

Clip length. Clip length depends on the model: up to 8 seconds on Veo 3.1, 15 on Kling 3.0, and 30 on Kling 4.0. Longer content still means chaining clips together and cutting between them in post.

Character consistency across clips. Each new generation starts from scratch. The same face in clip 1 will look slightly different in clip 2 unless you use an identity tool like Soul ID or reference images.

Precise motion control. Describing complex movement in text is still imprecise. “She reaches slowly toward the camera” will get interpreted differently each time. First-and-last-frame control, available in models like Wan 2.7, is the most direct solution, but it requires planning both endpoints upfront.

Multi-person scenes. Two people interacting in the same shot, especially in close-up, often produce visible inconsistencies. Generate each person separately and composite in post if precision matters.