AI-powered content material technology is now embedded in on a regular basis instruments like Adobe and Canva, with a slew of businesses and studios incorporating the expertise into their workflows. Picture fashions now ship photorealistic outcomes constantly, video fashions can generate lengthy and coherent clips, and each can comply with inventive instructions.
Creators are more and more working these workflows regionally on PCs to maintain property beneath direct management, take away cloud service prices and get rid of the friction of iteration — making it simpler to refine outputs on the tempo actual inventive tasks demand.
Since their inception, NVIDIA RTX PCs have been the system of selection for working inventive AI on account of their excessive efficiency — lowering iteration time — and the truth that customers can run fashions on them without spending a dime, eradicating token anxiousness.
With current RTX optimizations and new open-weight fashions launched at CES earlier this month, creatives can work sooner, extra effectively and with far better inventive management.
Easy methods to Get Began
Getting began with visible generative AI can really feel complicated and limiting. On-line AI mills are simple to make use of however supply restricted management.
Open supply group instruments like ComfyUI simplify establishing superior inventive workflows and are simple to put in. Additionally they present a simple strategy to obtain the newest and best fashions, resembling FLUX.2 and LTX-2, in addition to prime group workflows.
Right here’s easy methods to get began with visible generative AI regionally on RTX PCs utilizing ComfyUI and in style fashions:
- Go to comfortable.org to obtain and set up ComfyUI for Home windows.
- Launch ComfyUI.
- Create an preliminary picture utilizing the starter template:
-
- Click on on the “Templates” button, then on “Getting Began” and select “1.1 Starter – Textual content to Picture.”
- Join the mannequin “Node” to the “Save Picture Node.” The nodes work in a pipeline to generate content material utilizing AI.
- Press the blue “Run” button and watch the inexperienced “Node” spotlight because the RTX-powered PC generates its first picture.
Change the immediate and run it once more to enter extra deeply into the inventive world of visible generative AI.
Learn extra beneath on easy methods to dive into extra ComfyUI templates that use extra superior picture and video fashions.
Mannequin Sizes and GPUs
As customers get extra acquainted with ComfyUI and the fashions that assist it, they’ll want to contemplate GPU VRAM capability and whether or not a mannequin will match inside it. Listed here are some examples for getting began, relying on GPU VRAM:
Producing Photos
To discover easy methods to enhance picture technology high quality utilizing FLUX.2-Dev:
From the ComfyUI “Templates” part, click on on “All Templates” and seek for “FLUX.2 Dev Textual content to Picture.” Choose it, and ComfyUI will load the gathering of linked nodes, or “Workflow.”
FLUX.2-Dev has mannequin weights that can have to be downloaded.
Mannequin weights are the “data” inside an AI mannequin — consider them just like the synapses in a mind. When a picture technology mannequin like FLUX.2 was educated, it realized patterns from hundreds of thousands of photos. These patterns are saved as billions of numerical values known as “weights.”
ComfyUI doesn’t include these weights in-built. As a substitute, it downloads them on demand from repositories like Hugging Face. These recordsdata are massive (FLUX.2 may be >30GB relying on the model), which is why methods want sufficient storage and obtain time to seize them.
A dialog will seem to information customers by way of downloading the mannequin weights. The burden recordsdata (filename.safetensors) are routinely saved to the right ComfyUI folder on a person’s PC.
Saving Workflows:
Now that the mannequin weights are downloaded, the following step is to avoid wasting this newly downloaded template as a “Workflow.”
Customers can click on on the top-left hamburger menu (three traces) and select “Save.” The workflow is now saved within the person’s record of “Workflows” (press W to point out or cover the window). Shut the tab to exit the workflow with out dropping any work.
If the obtain dialog was unintentionally closed earlier than the mannequin weights completed downloading:
- Press W to rapidly open the “Workflows” window.
- Choose the Workflow and ComfyUI will load it. This may even immediate for any lacking mannequin weights to obtain.
- ComfyUI is now able to generate a picture utilizing FLUX.2-Dev.
Immediate Suggestions for FLUX.2-Dev:
- Begin with clear, concrete descriptions of the topic, setting, model and temper — for instance: “Cinematic closeup of a classic race automobile within the rain, neon reflections on moist asphalt, excessive distinction, 35mm pictures.” Brief‑to‑medium size prompts — a single, centered sentence or two — are often simpler to regulate than lengthy, storylike prompts, particularly when getting began.
- Add constraints to information consistency and high quality. Specify issues like:
- Framing (“huge shot” or “portrait”)
- Element degree (“excessive element, sharp focus”)
- Realism (“photorealistic” or “stylized illustration”)
- If outcomes are too busy, take away adjectives as an alternative of including extra.
- Keep away from destructive prompting — follow prompting what’s desired.
Study extra about FLUX.2 prompting in this information from Black Forest Labs.
Save Areas on Disk:
As soon as performed refining the picture, proper click on on “Save Picture Node” to open the picture in a browser, or put it aside in a brand new location.
ComfyUI’s default output folders are sometimes the next, primarily based on the applying sort and OS:
- Home windows (Standalone/Moveable Model): The folder is often present in C:ComfyUIoutput or an identical path inside the place this system was unzipped.
- Home windows (Desktop Utility): The trail is often positioned inside the AppData listing, like: C:UserspercentusernamepercentAppDataLocalPrograms@comfyorgcomfyui-electronresourcesComfyUIoutput
- Linux: The set up location defaults to ~/.config/ComfyUI.
Prompting Movies
Discover easy methods to enhance video technology high quality, utilizing the brand new LTX-2 mannequin for example:
Lightrick’s LTX‑2 is a complicated audio-video mannequin designed for controllable, storyboard-style video technology in ComfyUI. As soon as the LTX‑2 Picture to Video Template and mannequin weights are downloaded, begin by treating the immediate like a brief shot description, quite than a full film script.
Not like the primary two Templates, LTX‑2 Picture to Video combines a picture and a textual content immediate to generate video.
Customers can take one of many photos generated in FLUX.2-Dev and add a textual content immediate to present it life.
Immediate Suggestions for LTX‑2:
For greatest ends in ComfyUI, write a single flowing paragraph within the current tense or use a easy, script‑model format with scene headings (sluglines), motion, character names and dialogue. Purpose for 4 to 6 descriptive sentences that cowl all the important thing facets:
- Set up the shot and scene (huge/medium/closeup, lighting, colour, textures, ambiance).
- Describe the motion as a transparent sequence, outline characters with seen traits and physique language, and specify digital camera strikes.
- Lastly, add audio, resembling ambient sound, music and dialogue, utilizing citation marks.
- Match the extent of element to the shot scale. For instance, closeups want extra exact character and texture element than huge pictures. Be clear on how the digital camera pertains to the topic, not simply the place it strikes.
Further particulars to contemplate including to prompts:
- Digital camera motion language: Specify instructions like “sluggish dolly in,” “handheld monitoring,” “over‑the‑shoulder shot,” “pans throughout,” “tilts upward,” “pushes in,” “pulls again” or “static body.”
- Shot sorts: Specify huge, medium or shut‑ups with considerate lighting, shallow depth of subject and pure movement.
- Pacing: Direct for sluggish movement, time‑lapses, lingering pictures, steady pictures, freeze frames or seamless transitions that form rhythm and tone.
- Environment: Add particulars like fog, mist, rain, golden hour mild, reflections and wealthy floor textures that floor the scene.
- Model: Early within the immediate, specify kinds like painterly, movie noir, analog movie, cease‑movement, pixelated edges, vogue editorial or surreal.
- Lighting: Direct backlighting, particular colour palettes, comfortable rim mild, lens flares or different lighting particulars utilizing particular language.
- Feelings: Deal with prompting for single‑topic performances with clear facial expressions and small gestures.
- Voice and audio: Immediate characters to talk or sing in several languages, supported by clear ambient sound descriptions.
Optimizing VRAM Utilization and Picture High quality
As a frontier mannequin, LTX-2 makes use of vital quantities of video reminiscence (VRAM) to ship high quality outcomes. Reminiscence use goes up as decision, body charges, size or steps improve.
ComfyUI and NVIDIA have collaborated to optimize a weight streaming characteristic that enables customers to dump elements of the workflow to system reminiscence if their GPU runs out of VRAM — however this comes at a value in efficiency.
Relying on the GPU and use case, customers could wish to constrain these components to make sure affordable technology occasions.
LTX-2 is an extremely superior mannequin — however as with every mannequin, tweaking the settings has a big effect on high quality.
Study extra about optimizing LTX-2 utilization with RTX GPUs within the Fast Begin Information for LTX-2 In ComfyUI.
Constructing a Customized Workflow With FLUX.2-Dev and LTX-2
Customers can simplify the method of hopping between ComfyUI Workflows with FLUX.2-Dev to generate a picture, discovering it on disk and including it as a picture immediate to the LTX-2 Picture to Video Workflow by combining the fashions into a brand new workflow:
- Open the saved FLUX.2-Dev Textual content to Picture Workflow.
- Ctrl+left mouse click on the FLUX.2-Dev Textual content to Picture node.
- Within the LTX-2 Picture to Video Workflow, paste the node utilizing Ctrl+V.
- Merely hover over the FLUX.2-Dev Textual content to Picture node IMAGE dot, left click on and drag to the Resize Picture/Masks Enter dot. A blue connector will seem.
Save with a brand new title, and textual content immediate for picture and video in a single workflow.
Superior 3D Era
Past producing photos with FLUX.2 and movies with LTX‑2, the following step is including 3D steerage. The NVIDIA Blueprint for 3D-guided generative AI exhibits easy methods to use 3D scenes and property to drive extra controllable, production-style picture and video pipelines on RTX PCs — with ready-made workflows customers can examine, tweak and lengthen.
Creators can showcase their work, join with different customers and discover assistance on the Steady Diffusion subreddit and ComfyUI Discord.
#ICYMI — The Newest Developments in NVIDIA RTX AI PCs
💻NVIDIA @ CES 2026
CES bulletins included 4K AI video technology acceleration on PCs with LTX-2 and ComfyUI upgrades. Plus, main RTX accelerations throughout ComfyUI, LTX-2, Llama.cpp, Ollama, Hyperlink and extra unlock video, picture and textual content technology use instances on AI PCs.
📝 Black Forest Labs FLUX 2 Variants
FLUX.2 [klein] is a set of compact, ultrafast fashions that assist each picture technology and modifying, delivering state-of-the-art picture high quality. The fashions are accelerated by NVFP4 and NVFP8, boosting pace by as much as 2.5x and enabling them to run performantly throughout a variety of RTX GPUs.
✨Undertaking G-Help Replace
With a brand new “Reasoning Mode” enabled by default, Undertaking G-Help positive factors an accuracy and intelligence increase, in addition to the flexibility to motion a number of instructions directly. G-Help can now management settings on G-SYNC displays, CORSAIR peripherals and CORSAIR PC elements by way of iCUE — protecting lighting, profiles, efficiency and cooling.
Assist can be coming quickly to Elgato Stream Decks, bringing G-Help nearer to a unified AI interface for tuning and controlling almost any system. For G-Help plug-in devs, a brand new Cursor-based plug-in builder accelerates growth utilizing Cursor’s agentic coding surroundings.
Plug in to NVIDIA AI PC on Fb, Instagram, TikTok and X — and keep knowledgeable by subscribing to the RTX AI PC e-newsletter.
Observe NVIDIA Workstation on LinkedIn and X.
See discover relating to software program product data.
