LTX-2 Text to Video in ComfyUI: Generate Realistic AI Videos from Text Prompts

Prompt used

colorful, animated puppet character performing on a disco stage with a glowing disco ball overhead and moving colorful lights. The puppet has big button eyes, a wide fabric mouth, and playful, exaggerated movements. Lighting is bright, energetic, and fun.

The camera holds a medium shot as the puppet grabs a microphone and starts singing with clear lip-sync and bouncy, cartoon-style motion.

The puppet sings:
🎵 "Clap your hands, spin around,
Dancing feet won't touch the ground!
Shake it left and shake it right,
Puppet party every night!" 🎵

While singing, the puppet dances, wiggles its body, points the microphone toward the camera, and spins happily under the disco ball. Its eyes blink dramatically, its head bobs to the beat, and its mouth opens and closes in an exaggerated, comedic way.

The scene ends with the puppet striking a silly pose, holding the mic high as the disco lights flash.

Prompt used

A calm, cozy indoor scene with soft natural lighting and warm color tones. The camera shows a young woman sitting at a small wooden desk, gently tapping, brushing, or handling soft objects like paper, fabric, or brushes. Her movements are slow, deliberate, and soothing.

The camera alternates between close-ups of hands and face, showing realistic blinking, subtle head tilts, and gentle expressions. Ambient sounds like soft tapping, brushing, and light background music accompany the scene (for ASMR effect).

The overall atmosphere is relaxing and immersive, with slight cinematic micro-movements for realism. Background objects are softly blurred, with smooth depth of field, creating a cozy, peaceful vibe. Colors are warm and soft, no harsh contrasts, with natural textures.

The scene ends with the woman giving a gentle smile and looking at the camera, while continuing the slow, relaxing motions.

Prompt used

A realistic video of a confident, attractive young woman walking slowly through a bright flower garden in daylight. The scene is colorful and natural, with soft pink, white, yellow, and lavender flowers. Lighting is natural and even.

The camera tracks her in a medium shot as she walks forward, with a gently blurred floral background for depth. Her movement is smooth and realistic, with natural posture and arm motion.

She looks toward the camera and speaks clearly in natural English, with accurate lip-sync and facial movement. She says exactly:
"This video is created using LTX-2 Text to Video."
"This video is created using LTX-2 Text to Video."

She makes subtle hand gestures and natural facial expressions while talking. Hair and clothing move naturally as she walks.

The clip ends as she slows, faces the camera, and smiles confidently.

Prompt used

A photo-realistic 3D render of a human-like toy girl standing on a wooden table in soft daylight. She looks like a realistic collectible doll, with plastic skin texture, subtle surface imperfections, and detailed clothing fabric. Her facial features are gentle and expressive, blending realism with a toy-like design.

The camera begins in a medium shot and slowly moves closer, revealing fine details such as light reflections on her plastic skin, stitched clothing edges, and realistic shadows. Her head turns slightly, and she makes small, natural movements like blinking and a soft smile, giving the toy a lifelike presence.

The environment is simple and relatable — everyday objects like books or small props are visible in the background, slightly out of focus, reinforcing the scale of the toy. Lighting is soft and natural, with realistic reflections and depth of field.

The motion is slow and cinematic, emphasizing photo-real 3D quality, material realism, and experimental lifelike animation.

The scene ends with the toy girl looking toward the camera, holding a gentle pose as light softly wraps around her.

Update via the update folder

Go to your ComfyUI_windows_portable\update folder and double-click update_comfyui to update to the latest version.

Update all nodes via Manager

Launch ComfyUI, open the Manager, and click 'Update all' to update every installed custom node.

Switch to v0.8.1 or higher

Click Switch Version in ComfyUI. Ensure version is v0.8.1 or higher.

Set channel to dev / nightly

Confirm your channel is set to dev and the update version is ComfyUI nightly for the latest compatibility.

CKPT_name

Should be set to ltx-2-19b-dev-fp8.safetensors

LTXV audio text encoder loader

Should be set to gemma_3_12B_it_fp8_e4m3fn.safetensors

Enter your text prompt

Type a detailed description of the video you want to generate in the workflow's prompt box. Include motion, lighting, camera direction, and scene details for the best results.

Click Queue Prompt

Click 'Queue Prompt' to run the workflow. The generated video will be saved to your ComfyUI output folder automatically.

LTX-2 T2V is a generative AI model that creates smooth, realistic videos from text prompts inside ComfyUI. It supports cartoon animation, realistic footage, cinematic camera movement, and custom LoRA-based camera control.

You need: ltx-2-19b-dev-fp8.safetensors (checkpoint), gemma_3_12B_it_fp8_e4m3fn.safetensors (text encoder), ltx-2-19b-distilled-lora-384.safetensors and ltx-2-19b-lora-camera-control-dolly-left.safetensors (LoRAs), and ltx-2-spatial-upscaler-x2-1.0.safetensors (latent upscaler).

LTX-2 T2V requires ComfyUI v0.8.1 or higher. Use Switch Version in ComfyUI and set the channel to dev with ComfyUI nightly for the latest compatibility.

Open run_nvidia_gpu.bat in Notepad and add --reserve-vram 4 to the launch command. Lower this to 3, 2, or 1 on lower-end GPUs. You can also increase your Windows page file size for additional headroom.

Place the checkpoint in ComfyUI/models/checkpoints/, the text encoder in ComfyUI/models/text_encoders/, LoRA files in ComfyUI/models/loras/, and the upscaler in ComfyUI/models/latent_upscale_models/.