Sulphur 2 is an AI video generator built on LTX 2.3. You can create cinematic clips straight from your browser — no GPU, no downloads. Just write a prompt or drop in an image and you're done.
Run Sulphur 2 in your browser for instant results — or install locally via ComfyUI for unlimited generations and full workflow control.
Generate directly in your browser at sulphur2ai.com. Write a prompt, pick a format, and get your first cinematic clip in under 3 minutes. No software to install, no model weights to download.
Sulphur 2 is fully open source. Base weights are hosted at SulphurAI/Sulphur-2-base on Hugging Face — available in BF16, FP8 Mixed, and GGUF formats.
The fastest way to get a cinematic Sulphur 2 video — straight from your browser without any local setup.
Head to sulphur2ai.com and sign in. Once you're in, navigate to the Sulphur 2 generator. You'll see two generation modes at the top — Text-to-Video and Image-to-Video. Pick the one that matches your starting point.
New accounts come with starter credits. No credit card is required to get your first generation done.
Pro tip: Bookmark the generator URL directly — it saves you a click every session.
For Text-to-Video: write a 40–80 word shot description. Cover the subject, what it's doing, how the camera moves, the lighting, and the visual style. Sulphur 2 responds well to cinematic language — think director's notes, not casual descriptions.
For Image-to-Video: upload your reference image, then describe the motion you want — camera direction, subject movement, lighting shift. Don't re-describe the image; describe what should happen to it.
Key insight: Vague prompts produce average results. Get specific with camera and lighting terms — something like "slow dolly-in, soft blue rim light" makes a real difference in what you get out.
Pick the format that matches where your video will live. 16:9 landscape works for websites, YouTube, product pages, and desktop presentations. 9:16 vertical is built for TikTok, Instagram Reels, YouTube Shorts, and mobile ads.
Start at 720p for your first test — it's faster and lets you validate motion, framing, and style before committing to a full 1080p render. Once you're happy with the prompt, upgrade the resolution for a crisp final output.
Workflow tip: Always prototype at 720p. A single 1080p render costs more credits — validate first, upscale second. See credit packs before upgrading resolution.
Hit generate and let Sulphur 2 work. Online generations typically complete in 30 seconds to 3 minutes depending on resolution and server load. Once the clip renders, preview it in full — check the motion timing, subject consistency, camera movement, and overall look.
If the motion is too subtle, push the action language harder. If the style drifts, add more explicit style anchors. Even small tweaks to your wording can change the output a lot. Download your final clip once it's dialed in.
Iteration rule: Change one element at a time — camera, lighting, or style — so you know exactly what moved the needle. Changing everything at once makes it impossible to learn what worked.
Sulphur 2 responds better to structured, shot-style prompts than most other AI video tools. Think of your prompt as a compact director's shot description — subject + motion + camera + light + mood.
The base weights are publicly available at SulphurAI/Sulphur-2-base on Hugging Face. Researchers and developers can run locally via ComfyUI. Most users generate directly on sulphur2ai.com — no install required.
GGUF quantizations also available on the SulphurAI/Sulphur-2-base HuggingFace page for CPU/offload workflows.
Everything you need to know before your first generation. Updated based on what real users ask most.
Sulphur 2 is an AI video generator built on LTX 2.3. It creates short cinematic clips — up to 5 seconds — from text prompts or reference images. Run it free in your browser at sulphur2ai.com, or download the open weights from Hugging Face.
No GPU is needed for online generation at sulphur2ai.com — everything runs in your browser. For local ComfyUI use, aim for 16 GB+ VRAM (RTX 3080 Ti or better). The FP8 Mixed variant balances quality and speed at 16–24 GB.
Online clips typically finish in 30 seconds to 3 minutes, depending on resolution and server load. On a 24 GB GPU with FP8 Mixed at 720p, expect roughly 2–4 minutes per 5-second clip.
Yes — videos made on sulphur2ai.com can be used commercially under our Terms of Service. For self-hosted open weights, check the license on the Hugging Face model card. This is not legal advice.
The base (Dev) model delivers the highest quality but runs slower. The distilled LoRA variant is 40–60% faster with a modest quality trade-off. Use base for finals; use distilled for fast iteration.
Sulphur 2 accepts text prompts for text-to-video and reference images for image-to-video. GGUF quantizations on Hugging Face also support CPU and offload workflows. Audio input is not supported.
No. Your prompts and uploads are never used to train or fine-tune any model. They are processed only for your generation request. See our Privacy Policy for retention details.
Official ComfyUI workflow JSON files live in the SulphurAI GitHub repository. Four workflows cover text-to-video and image-to-video in base and distilled modes. Drag any JSON into ComfyUI to load it.