How I built a Claude AI workflow that generates images and video on demand
Full workflow & Seedance 2.0 chatter →
Welcome back to Visually AI!
Today’s reading time is 5 minutes.
In today’s edition:
Creating AI images and video inside VS Code using Claude Code
Seedance 2.0 in the USA
Image & video prompts
Recent top tool picks of the week
Creating Images & Video with Claude Code Inside VS Code
Most AI visual workflows look the same: open a browser, navigate to the tool, type a prompt, wait, download, repeat. And this is me, but with the development of Claude Code and agents I have been exploring other ways to keep the workflows streamlined and in one place.
Enter Claude Code via Anthropic’s AI extension for VS Code. It lets you run AI workflows directly inside your editor including generating images and video without ever leaving the window. It even downloads the result straight into your folder.
Workflow:
Step 1: Set up Claude Code in VS Code
Install the Claude Code extension from the VS Code marketplace. Connect your Anthropic account.
(I would recommend you take time to set up your Claude Code properly so it knows as much about you as possible if this is your first time!)
Step 2: Build your prompt-gen skill
This is where it gets interesting. Inside Claude Code, you can build your custom skills (or install one straight from GitHub). I built one called /prompt-gen.
When I run it, I give it a raw idea: “cinematic shot of a neon-lit alleyway, moody, Blade Runner feel” and it engineers a fully structured, tool-optimized prompt for whatever generator I’m targeting. Kling, Midjourney, Runway, Veo — the skill knows how each tool thinks and writes accordingly.
The prompt is ready to use. I can add as much input or direction and the skill knows my prompting style.
Step 3: Generate via kie.ai
Now we need API access across video and image tools. For this, I route everything through kie.ai. I also use other platforms such as Fal.ai. Kie has 66 models on its platform at good pricing including the latest releases (Seedance 2.0 / Wan 2.7).
Inside Claude Code, I set up another skill called /ai-generate. I input my kie.ai API credentials once for each specific model I want to use. From that point, Claude handles everything. I approve the prompt from /prompt-gen, run it through /ai-generate and the result downloads straight into my project folder on my desktop.
No browser. No downloading.
Step 4: Review inside VS Code
The generated asset also lands directly in my output folder; visible in the VS Code file tree. I can review it, iterate on the prompt, and regenerate; all without leaving the editor.
That’s the full workflow: idea → engineered prompt → API call → asset in folder (I have separated Image & Video). It will also be available in the cloud on kie.ai for up to 14 days.
I do not always use this workflow, but Claude Code inside VS Code is a strong Executive Assistant and knows a lot about me. By adding image & video it has helped with some of my projects, especially quick iteration and finding what works.
I would stress this is not my preferred workflow, but a useful asset to have!
This Seedance 2.0 video was generated through kie AI in VS Code without leaving VS Code:
Seedance 2.0 launch on multiple platforms
Seedance 2.0 has been around for a while, but it’s now available across more platforms and more widely accessible in the USA. If you haven’t tested it yet, here’s what’s actually worth knowing.
What it offers
The headline upgrade over previous versions is multi-modal input. Where older Seedance models were straightforward text-to-video and image-to-video, 2.0 lets you combine images, video references, and audio in a single generation. You can reference motion, camera movement, and visual style from existing footage and apply it to new content; without having to describe every detail in the prompt.
Video extension is also a meaningful addition. You can extend an existing clip while maintaining continuity rather than regenerating from scratch. And it handles complex, layered prompts faster than most models currently available.
The reality
Seedance 2.0 took significant backlash from Hollywood early on with concerns around realistic human likeness generation led to capability restrictions that are still in place. In practice, this means human face references get rejected consistently. If your workflow relies on real human subjects, you’ll hit that wall fast.
For that reason, it’s not yet a replacement for Kling 3.0 in most creative workflows. Where Kling gives you more freedom, Seedance 2.0 earns its place in motion replication, reference-based generation, and non-human subject work.
Worth having in the toolkit. Just know where it fits.
Here’s a short list of platforms you can find Seedance 2.0 now:
And many more…
Here is a Seedance 2.0 Product Ad Workflow on Lovart AI in action with a single reference image:
🖼️ Image Prompts
Prompt: Neon sign on top of a trendy urban cafe, reading “Crave” in neon blue futuristic serif font reflected in wet asphalt at 2am, slight rain still falling, no people, bokeh traffic light in distance, desaturated urban tones
Prompt: Ultra-realistic 9-panel cinematic storyboard grid, 3 rows x 3 columns, 16:9 aspect ratio. Each panel tells one chapter of a sequential futuristic narrative featuring the same man throughout: late 30s, close-cropped dark hair, angular jaw, visible stubble, dark graphite tactical jacket with matte black collar, utility trousers. Consistent character across all nine panels.
Panel 1: wide shot, man stepping off a transit pod onto a rain-slicked platform, vertical city towers behind, teal and amber holographic signage reflecting off wet concrete. Panel 2: over-the-shoulder, pushing through a crowded elevated street market, holographic vendor displays, warm amber food stall light. Panel 3: medium close-up in a narrow alley, receiving a translucent data chip from a hooded figure’s hand, faded holographic graffiti on walls. Panel 4: close-up of hands holding the data chip projecting a faint amber holographic map, face lit from below. Panel 5: portrait at a massive observation window, layered city below in teal and amber, his reflection faint in the glass. Panel 6: low angle, descending an industrial staircase into the lower city, flickering amber utility lights, steam at ankle height. Panel 7: wide-medium, standing before a massive sealed door underground, teal light pulsing along its seams, data chip in hand. Panel 8: stepping into a vast hidden server hall, towering server columns emitting cold teal glow, amber warning lights at floor level. Panel 9: close-up of hand inserting the data chip into a console, screen igniting with cascading amber data streams.
Consistent color palette across all panels: teal, gunmetal, warm amber accents. 35mm lens, f/2.0, ISO 400 simulation throughout. Thin white borders separating each panel. Cinematic color grade, anamorphic framing. Natural skin texture with visible pores. Do not beautify or alter facial features.
Negative: plastic skin, skin smoothing, beautification filters, anatomy normalization, extra fingers, CGI, oversaturated colors, cartoon, dataset-average proportions, blurry, low resolution, distorted face
🎥 Video Prompts
Prompt: Premium brand reveal for a digital newsletter called “Visually AI.” Pure black background. 0-2s: Frame starts in darkness, a single cool-white light slowly rises, illuminating floating 3D glass letters spelling “Visually AI” that rotate gently with ray-traced reflections. 3-5s: Camera pushes through the letters into a cascade of AI-generated artwork — fashion editorials, cinematic stills, product shots — floating as translucent holographic cards that drift and rotate in space. 6-8s: The camera weaves between the floating cards as they arrange themselves into a grid, each card pulsing with subtle light. 9-10s: Camera pulls back smoothly to reveal the full grid forming the shape of a glowing display, with “Visually AI” centered beneath it. Volumetric lighting, glass refractions, dark premium aesthetic, 4K, shallow depth of field, smooth continuous camera movement, cinematic electronic score. [REFERENCE LOGO ATTACHED]
→ Seedance 2.0 on Lovart AI
Prompt: A continuous drone shot begins high above a vast glacial lake at dawn. The camera descends rapidly toward the mirror-still water surface, pulling up at the last moment to skim inches above the reflection. It accelerates toward a lone figure standing on a rock outcrop at the lake's edge. As the camera reaches them, it spirals upward in a tight corkscrew, revealing the full scale of snow-capped mountains surrounding the valley. Morning mist rolls through the peaks. The camera completes the spiral and locks into a wide hero frame — the figure small against the vast landscape, golden sunrise breaking over the ridge. 4K hyper-realistic, volumetric fog, natural color grade, shallow atmosphere haze, epic orchestral swell.
→ Seedance 2.0 on Lovart AI
🎨 Explore my Portfolio
I’ve had some amazing opportunities to work on a variety of projects recently, and I don’t share everything in this newsletter or online. And you know I love testing tools to share.
Take a look at my portfolio to get a quick glimpse of my work:
🚀 My Recent Top AI Tools Picks
Claude Code extension for VS Code — AI assistant extension for your Visual Studio Code. Handles prompts, workflows, and generation without leaving your editor.
kie.ai — Single API layer for 66+ models including Nano Banana, Kling, Seedance, Runway, and Wan. Competitive pricing.
Hermes Agent - A tool-using AI assistant that can research, write, automate, and execute complex workflows across files, the web, and connected apps.
World Labs Spark 2.0 - Render 3D Guassian Splatting (3DGS) in your web browser. Incorporates a new Level-of-Detail (LoD) system that can stream and render huge 3DGS worlds on any device.
Thank you for reading.
I hope you have a creative week!
Heather Cooper