The Kandinsky 6.0 prompt generator gives you 20 free, copy-paste prompts for the MIT-licensed open-source model that generates synchronized video and audio in one pass. Waves with real crash sounds. Music that matches the instrument on screen. Audio locked to every visual event — no post-production required.
The Kandinsky 6.0 prompt generator on this page provides 20 free, professionally crafted prompts for Kandinsky 6.0, the open-source AI video model released by Kandinsky Lab on October 6, 2026. Kandinsky 6.0 is MIT-licensed and generates synchronized video and audio in a single generation pass — the model creates both the visual content and the corresponding soundscape simultaneously, with audio events locked to their visual causes.
Most AI video models generate silent video. Kandinsky 6.0 changes this: it produces 5-second clips with 44 kHz synchronized audio, including lip-sync, in a single pass. A hammer strike produces a ring. A wave crash produces a crash. A piano key produces a note in the acoustic space shown. Two model variants serve different hardware contexts: a 29B Pro for maximum quality and a 3B Lite that runs on a consumer RTX 4090 — making synchronized audio-video generation available to individual creators without cloud infrastructure for the first time.
Every prompt below is structured for Kandinsky 6.0's audio-visual synthesis: each describes both the visual scene and the expected soundscape explicitly, with timing cues and production quality references. Paste directly into Kandinsky 6.0 via ComfyUI, HuggingFace Spaces, or the kandinskylab API.
Kandinsky 6.0 responds directly to explicit audio description in your prompt. Use this six-part structure:
Click any prompt to copy — paste into Kandinsky 6.0 via ComfyUI, HuggingFace Spaces, or the kandinskylab API
Aerial drone footage descending toward an active volcanic summit at dusk: lava flows visible on three flanks, a continuous column of ash and steam rising 200 metres. The sky is deep red-orange behind the plume. Audio: a deep low-frequency ground rumble sustained throughout, punctuated by sharp percussive explosions as gas vents release — each explosion precisely synchronized with a visible pressure burst in the ash column. The audio grows louder as the drone descends. BBC Earth quality. 5 seconds.
A medium shot inside a dimly lit piano bar at midnight: a pianist in a white shirt plays an improvised jazz passage alone on a baby grand, the keys reflecting a single overhead spotlight. Audio: the piano fills the room — finger pressure translates directly into note volume, sustain pedal audible as a soft click, the room's natural reverb adding warmth, a background hum of conversation barely audible. Every key press is locked to its sound. Monocle-quality atmosphere. 5 seconds.
An extreme close-up of water cascading over a moss-covered granite ledge in a Scottish mountain stream: each water thread visible as an individual flow, spray catching the late afternoon light in rainbow fragments. Audio: the micro-scale crash of water against rock — high-frequency hiss of the spray, mid-frequency rush of the main flow, a bubbling below. The audio is spatially locked: sounds emerge from their exact visual position. Natural World documentary quality. 5 seconds.
A walking-pace shot through a Bangkok night market at 10 PM: neon signs reflecting off rain-wet pavement, a cook at a wok station flipping ingredients, motorbikes threading between stalls, incense smoke drifting. Audio: the wok sizzle and flame roar at the food stall, motorbike engines spatially placed as they pass, Thai vendor calls, the collective hum of the market. Sound sources move with their visual positions as the camera walks through. 5 seconds.
A static wide shot of an open countryside road during an electrical storm: lightning strikes directly overhead — a white flash filling the entire frame — followed immediately by a thunderclap that physically shakes the camera. Rain is falling hard; wind bends the roadside trees. Audio: the lightning's electrical crack arrives simultaneously with the visual flash, then the full chest-impact thunderclap, then a 3-second rolling echo off distant hills. Rain and wind continuous underneath. 5 seconds.
Close medium shot of a blacksmith hammering a glowing iron rod at an open forge: each hammer strike produces a shower of sparks and a ringing impact. The forge glows deep orange. The smith's face is focused, lit only by the forge light. Audio: the hammer strike is the anchor — each impact produces a sharp metallic ring followed by a brief resonance tail. Sparks visible in the air are silent; the ring arrives on contact. The bellows breathe in the background between strikes. 5 seconds.
A single cellist performs an unaccompanied Bach suite in an empty medieval stone church: the bow draws long notes on the lower strings, the camera holds at a medium distance, the cello and player silhouetted against tall stained glass. Audio: the cello tone fills the stone space with 2.5 seconds of reverb, the bow hair on string audible as a soft scratch beneath the tone, the resonance of the body visible in the wood grain. The acoustic space of the church makes the sound come alive. 5 seconds.
Interior of a WWII submarine control room at battle stations: red light, officers at stations, the sonar operator with headphones listening intently. The sonar pings once — a high-pitched sustained ring that slowly fades. Audio: the sonar ping is clean and precise against an otherwise near-silent room. Breathing, the creak of the hull under pressure, the faint hiss of air circulation. The silence after the ping feels loaded. Period-accurate reconstruction. 5 seconds.
A chase-mounted camera on a motorcycle accelerating through a desert canyon at golden hour: terracotta canyon walls blur past, the road curves left then right, dust rises in the wake. Audio: the motorcycle engine — starting in its mid-range working tone as the canyon entrance is reached, then a rising throttle note as the rider accelerates through a straight, the sound bouncing off the canyon walls and returning as a distinct echo 0.3 seconds later. Wind in the mic at speed. 5 seconds.
A Venetian glassblower gathers molten glass from a furnace on a long iron pipe, then blows gently — the glass bubble expands visibly with each breath. The workshop is warm amber. Audio: the crackling roar of the furnace behind, the soft human exhalation through the pipe amplified by its length into a bass resonance in the tube, the gentle creak of the iron pipe rotating in the blower's hands. Silence except for craft sounds. 5 seconds.
Close-up of a corrugated iron roof during a tropical downpour: each raindrop impact visible as a tiny crown of water, the surface vibrating at the highest rainfall density zones. Audio: the corrugated iron transforms rain into a percussive instrument — individual impacts audible in the quieter edge zones, dense impacts merge into a sustained roar at the centre, metallic resonance adds harmonic richness. The rhythm shifts as gusts intensify. 5 seconds.
A bear standing in a shallow salmon river catches a leaping fish in its mouth at frame centre: the fish jumps, the bear lunges, impact. Water explodes outward in a perfect arc. Audio: the water splashes are synchronized precisely to every movement — bear's entry splash, the fish breaking the surface, the catch impact and secondary splash. The river's constant rush underneath. No music. BBC Planet Earth quality. 5 seconds.
Night street: a large neon sign in a Tokyo alley flickers — on, off, on — casting its reflection on the rain-wet cobblestones below in a 1-second fault cycle. Audio: the neon's electrical buzz is heard — a sustained 60Hz hum with a sharp crackle on each flicker, the flicker sound and visual perfectly synchronized. Rain provides a soft continuous white noise bed. The effect is atmospheric and slightly unsettling. 5 seconds.
A shot from inside the gondola looking up: the burner fires — a column of flame 2 metres tall roaring into the balloon envelope above, the orange fire reflected in the interior silk. The burner cuts, silence. Audio: the burner blast is the event — a roaring jet of flame that fills the audio spectrum, then instant silence when the valve closes. Wind audible in the silence after. The heat distortion visible in the air above the flame during the blast. 5 seconds.
A single dancer on a stage lit from the wings performs an arabesque that flows into a grand jeté: the arc of the jump perfectly framed against a black backdrop, pointe shoes visible as she lands with controlled force. Audio: the hardwood stage floor under pointe shoes — the soft thud of preparation, the near-silence at the peak of the jump, the controlled impact landing at the 3-second mark. An auditorium's silence amplifies every sound. 5 seconds.
Inside a glacial ice cave: deep blue light filters through 10 metres of ice above, ice crystals hang from the ceiling, the walls luminescent azure. A drip of meltwater falls every 2 seconds. Audio: the cave's profound silence is the canvas — the single water drip rings clearly in the acoustic space, its echo lasting 1.2 seconds in the ice chamber. Distant deep structural groaning of the glacier. No artificial sound. 5 seconds.
A violinist performs in a New York subway corridor: the tile walls create a natural reverb chamber, commuters pass in both directions, some slowing briefly. Audio: the violin has the corridor's distinctive wet acoustic — the overtones bounce back 0.4 seconds after the bow, creating a natural layered wash. Subway rumble enters from behind at the 3-second mark, a low-frequency wall of sound that makes the violin cut through even harder. 5 seconds.
Industrial foundry: a ladle tips and a stream of white-hot molten metal pours into a mould. The pour streams in slow motion for 2 seconds then cuts to the solidifying glow. Audio: the pour produces a rushing hiss — liquid metal at 1500°C vaporises contaminants audibly, a sustained white noise with occasional sharp pops as inclusions release. The mould creaks as it receives the weight and thermal expansion. No music. Documentary authenticity. 5 seconds.
A domestic cat on a windowsill in morning light stretches into a full arch, yawns — jaw opening fully — then turns to look directly at the camera. Dust motes visible in the beam of light. Audio: the cat's yawn produces a soft vocalization — a high-pitched breathy squeak, then the click of the jaw closing, the purr resuming immediately after. Window glass resonates faintly with street noise below. 5 seconds of pure warmth.
Slow time-lapse compressed to 5 seconds: a large radio telescope array in a high desert rotates together in coordinated tracking — 20 dishes turning in synchronized arcs against a star-filled sky, the Milky Way visible above them. Audio: the machinery of the dish drive motors — a low mechanical hum, gear engagement audible as a series of soft clicks, then the satisfied settling tone as each dish locks onto its target. Night wind across the array. 5 seconds.
Kandinsky 6.0 is the only open-source model that combines synchronized audio, consumer GPU inference, and MIT licensing:
| Model | Native Audio | Consumer GPU | Licence | Best For |
|---|---|---|---|---|
| Kandinsky 6.0 (Kandinsky Lab) ★ | Yes — 44 kHz sync | Yes (Lite: RTX 4090) | MIT | Audio+video sync, local inference, commercial use |
| SkyReels V4 (SkyWork AI) | Yes — #1 T2V-audio | High-VRAM only | Permissive | Highest audio sync quality, cloud API |
| LTX 2.5 (Lightricks) | No (silent video) | Yes — 22B efficient | Custom | Fast local video, #2 global leaderboard |
| Wan 3.0 (Alibaba Tongyi) | No (silent video) | High-VRAM preferred | Apache 2.0 | Thinking Mode, fine-tuning, research |
| Kling 4.0 (Kuaishou) | Partial (stereo) | No (API only) | Proprietary | 30-second 4K, multi-keyframe, cinematic |
★ Kandinsky 6.0 released October 6, 2026. Lite variant (3B) runs on consumer RTX 4090. Pro variant (29B) outperforms LTX 2.5 in visual and speech quality per Kandinsky Lab benchmarks. MIT-licensed for commercial use.
The Kandinsky 6.0 prompt generator on this page provides 20 free, copy-paste prompts for Kandinsky 6.0, the open-source AI video and audio generation model released by Kandinsky Lab on October 6, 2026. Kandinsky 6.0 generates synchronized video and audio in a single generation pass — the model produces both the visual content and the corresponding soundscape simultaneously, with audio locked to visual events. It is available under the MIT licence, making it free for commercial and research use.
Kandinsky 6.0 is an open-source AI model from Kandinsky Lab that generates synchronized video and audio in one pass. It comes in two variants: a 29B parameter Pro version for maximum quality and a 3B Lite version that can run on consumer GPUs such as an RTX 4090. Both variants generate 5-second clips with 44 kHz synchronized audio, support text-to-video and image-to-video modes, and can produce Full HD output (1920×1080) via a separate super-resolution upscale. The model is MIT-licensed and available on HuggingFace under the kandinskylab organization, with first-party ComfyUI workflow templates released on the same day.
Both Kandinsky 6.0 and SkyReels V4 generate synchronized video and audio in a single pass — a capability that most AI video models (Kling 4.0, Veo 4, HappyHorse) do not offer. The key difference is hardware access: Kandinsky 6.0 Lite runs on a consumer RTX 4090, while SkyReels V4 requires enterprise-class GPU infrastructure for full quality. Kandinsky 6.0 Pro outperforms LTX 2.5 in visual and speech quality according to the model's own benchmark. For creators who want synchronized audio+video with local inference on consumer hardware, Kandinsky 6.0 Lite is currently the most accessible option. For cloud inference without hardware constraints, both are available via APIs.
Kandinsky 6.0 prompts work best when they explicitly describe both the visual scene and the audio environment. A reliable structure: (1) Camera type and position — 'static wide-angle,' 'close medium shot,' 'aerial drone'; (2) Visual subject and action — physical, concrete description of what happens; (3) Audio description — name specific sound sources, describe their volume relationship, and note any timing cues (e.g., 'audio arrives 0.3 seconds after the visual'); (4) Environment and lighting — time of day, weather, light quality; (5) Production quality reference — 'BBC Earth,' 'Dolby Atmos spatial,' 'documentary authenticity'; (6) Duration — always end with '5 seconds' since that is Kandinsky 6.0's native generation length.
Kandinsky 6.0 model weights are available on HuggingFace under the kandinskylab organization. The Lite variant can be run locally on a consumer GPU (RTX 4090, 24GB VRAM). The Pro variant requires higher-end infrastructure — cloud inference via API is recommended. ComfyUI workflow templates shipped on the same day as the model, making it immediately usable within the ComfyUI ecosystem without custom node development. The MIT licence allows commercial use, derivative models, and fine-tuning without royalty payments.
Kandinsky 6.0 Pro is a 29B parameter model trained for maximum visual and audio quality — it produces better physics simulation, more precise audio synchronization, and higher coherence in complex multi-source audio scenes. Kandinsky 6.0 Lite is a 3B parameter model optimized for consumer GPU inference — it sacrifices some fine detail and audio complexity in exchange for running on hardware most creators already own (RTX 4090 or equivalent). For most social media and content creation use cases, Lite is sufficient. For cinematic or documentary quality output, Pro is the right choice, accessed via cloud API rather than locally.
First open-source video+audio model — #1 T2V-with-audio leaderboard
Lightricks 22B open-source video — fast local inference
Alibaba open-source video with Thinking Mode
30-second 4K video, up to 10 keyframes, cinematic quality
AI music and audio generation — up to 6 minutes
AI music generation — 20 genre prompts, Suno V5.5