Stay local
Run generation on your own compatible hardware with local model, VAE, LoRA, face-reference and output folders. After activation, there is no per-image 65West generation charge in the creative loop.
Private workflow
65WEST AI IMAGE GENERATOR (ZT/K2)
Your GPU.
Your models.
Your creative control.
65West AI Image Generator (ZT/K2) brings compatible Z-Image Turbo and Krea 2 models into one focused creative workstation. Switch model families with family-aware settings and VAE handling, generate and compare compatible LoRAs, refine detail and lighting, use face tools, upscale, and present your work without a per-image 65West cloud charge.
$39.95 USD
One-time purchase
Two-device license
LOCAL-FIRSTNo per-image cloud feeUse your own compatible model and GPU.
ONE WORKSPACEFrom prompt to historyGeneration controls, LoRAs, monitoring and review stay together.
1.5×–4×Optional model upscaling
LoRAStack + direct strength editing
1.25× / 1.50×Second-pass refinement
Face toolsReference + restoration
LiveSampling preview
View full size ↗
CREATED WITH 65WEST
Real output.
Generated locally.
This gallery features genuine images created with 65West AI Image Generator—not illustrative stand-ins. Each sample demonstrates the creative results possible within the application’s local model workflows.
864 × 1152Original resolution
3:4Portrait format
LocalGPU generation
Open any image to inspect the original full-resolution sample. Every gallery entry is genuine application output selected for the website.
02 · ENVIRONMENT
Tropical coast at golden hour
A wide environmental study with layered foliage, mist, shoreline detail, reflective water and natural light.
1152 × 864Original resolution
4:3Landscape format
View full size ↗
View full size ↗
03 · CINEMATIC CHARACTER
Stormlight above the city
A dramatic character scene balancing a clear action pose, reflective rain detail, layered architecture and high-contrast lightning.
1024 × 1024Original resolution
1:1Square format
04 · AUTOMOTIVE ENVIRONMENT
Classic steel under city rain
A cinematic 1965 Ford Mustang study combining wet-road reflections, brick textures, alley depth, atmospheric steam and contrasting cool and warm light.
1152 × 864Original resolution
4:3Landscape format
Genuine outputGenerated with the application
View full size ↗
View full size ↗
05 · INTERIOR DESIGN
One frame. An entire designed space.
A wide presidential-suite concept with connected living, dining, sleeping and hospitality zones. The image balances architectural scale with detailed stone, wood, marble, upholstery, glass and warm integrated lighting.
1152 × 864Original resolution
4:3Landscape format
Interior conceptSpace, materials and lighting
06 · WILDLIFE PORTRAIT
Winter stillness. Fine natural detail.
A square wildlife study centered on a white wolf with pale blue eyes. Layered fur, clinging snow, subtle tonal variation and a softly defocused forest keep the subject crisp without flattening the environment.
1024 × 1024Original resolution
1:1Square format
Wildlife detailFur, snow and depth of field
View full size ↗
View full size ↗
07 · STYLIZED CHARACTER
A different visual language.
A polished animation-inspired character scene with an expressive central pose, clean stylized rendering and a fully developed gaming environment. Warm window light meets cyan and violet accents to give the room depth, energy and a strong visual identity.
864 × 1152Original resolution
3:4Portrait format
Stylized sceneCharacter, environment and color
08 · PRODUCT ADVERTISING
Built to hold the spotlight.
A premium wine-advertising concept staged inside a stone cellar at sunset. Reflective glass, polished metal, natural wood, fruit and vine details create a convincing commercial composition, while the vineyard beyond adds atmosphere and a clear sense of place.
864 × 1152Original resolution
3:4Portrait format
Commercial conceptProduct, materials and lighting
View full size ↗
09 · FANTASY ENVIRONMENT
Build an entire world.
A cinematic mountain kingdom built around a towering castle, winding river and long stone approach. Layers of forest, mist, waterfalls and snow-covered peaks create scale and depth, while the break in the storm directs warm light toward the architectural focal point.
View full size ↗
1152 × 864Original resolution
4:3Landscape format
World-buildingArchitecture, atmosphere and scale
View full size ↗
10 · CULINARY PHOTOGRAPHY
Every detail adds flavor.
A restaurant-quality Thai curry presentation built around color, texture and atmosphere. Rising steam reinforces the sense of a freshly served dish, while polished wood, brass tableware, carved interiors and a moonlit tropical garden turn the meal into a complete hospitality scene.
1152 × 864Original resolution
4:3Landscape format
Culinary conceptFood, setting and atmosphere
A FOCUSED LOCAL WORKFLOW
Powerful enough to grow with you.
Focused enough to actually use.
Skip the node graph
The controls that matter for Z-Image Turbo and Krea 2 live in one desktop workspace: model family, prompt, dimensions, seed, sampler, compatible LoRAs, refinement, face tools, post processing, upscaling, history and viewing.
Purpose-built UI
Test with control
Reuse the last cached latent and generation state while changing the active LoRA stack. Compare a style or strength against the same underlying generation instead of guessing.
Controlled comparisons
MORE THAN “TYPE A PROMPT”
One studio for Z-Image Turbo and Krea 2
From first idea to refined output, 65West keeps the work inside one session.
Model-family switching
Move between compatible Z-Image Turbo and Krea 2 models with family-aware defaults, scheduler behavior, automatic VAE selection, and a deep pipeline reset when changing families.
Purpose-built generation controls
Use recommended 3:4 portrait and 4:3 landscape presets or set custom dimensions, then control steps, fixed or random seeds, sampler recipes, and scheduler shift.
Auto-Generate
Keep generating new variations continuously. Pair it with random-seed mode for an ongoing stream of fresh results until you stop.
LoRA stack management
Add compatible LoRAs, edit strengths directly, and independently disable Included LoRAs, ignore prompt-tagged LoRAs, or pause every LoRA without deleting your setup.
Wildcards + inline choices
Build reusable prompt variation with wildcard text files and nested inline alternatives for clothing, lighting, locations, styles and more.
Real second-pass refinement
Send the completed generation through a genuine image-to-image diffusion pass using the already-loaded model components—then return it to your chosen final size.
Face reference + Face Swap
Choose a source identity manually or call it directly from the prompt with a <face_ref:...> tag. Source analysis is cached for repeated use.
Automatic Face Detailer
Run GFPGAN restoration after generation and Face Swap, with a fidelity control that balances restoration strength against identity preservation.
Generation History
Review session thumbnails, inspect the generation prompt and metadata, save selected full-size images, identify already-saved results, and avoid accidental loss on exit.
Movable live preview
Watch sampling in Slow, Normal, Fast, or Real-Time mode. Move the preview anywhere over the workspace and let the application remember its position.
Adaptive post processing
Finish with optional tone, contrast, and brightness correction, saturation or grayscale controls, and repeatable seed-based film grain.
Model-based final upscaling
Use a compatible local PyTorch super-resolution model from 1.5× through 4.0×. Memory-safe tiling protects the completed normal-size image if upscaling cannot finish.
Viewing and presentation
Inspect an image at actual size in a frameless viewer with on-demand scrollbars, or turn the session into a fullscreen slideshow.
See the slideshow viewer ↓SECOND-PASS DIFFUSION, NOT A RESIZE
Give a strong generation a deliberate second pass.
Image Refinement reuses the already-loaded compatible model components for an image-to-image stage that can improve detail, lighting, tonal separation and depth. Memory is checked before the pass, and the program can automatically step down from 1.50× to 1.25× when headroom is unsafe. Optional final super-resolution upscaling is a separate finishing stage.
Analog PhotoFine DetailBalanced Detail & LightingNatural LightingStudio LightingCinematic LightingDramatic LightingGolden Hour
Open full size ↗
ORIGINAL GENERATIONREFINED RESULT
Drag the divider to inspect the genuine second-pass result.
ONE SESSION, SEVEN CREATIVE MODES
Explore the workflow
01 · PROMPT WITH STRUCTURE
Write what should be visible.
Build prompts around subject, appearance, action, environment, composition, lighting and medium. Add inline choices, wildcard tags, LoRA tags or a face-reference tag when you need variation or identity control.
A cinematic explorer portrait, warm rim lighting, detailed jungle temple, 35mm film character
<face_ref:Sarah>
__lighting__
<lora:film:0.8>
{sunrise|night}
LOCAL FACE SWAP WORKFLOW
Bring a permitted face reference into the generation.
Select a source manually or call it from the prompt with a face-reference tag. The application analyzes and caches the source, then replaces the largest detected face without sending the image to a 65West cloud service.
View source ↗
Local processingNo image is uploaded to a 65West cloud service.
Prompt or manual selectionUse a face-reference tag or browse directly to a permitted image.
Identity-focused comparisonThe slider uses the same generated image before and after Face Swap.
AFTER FACE SWAPORIGINAL GENERATION
Drag the divider across the portrait to inspect the genuine Face Swap result.
AUTOMATIC FACE DETAILER
Restore facial detail as part of the same local workflow.
The optional GFPGAN Face Detailer runs after generation and any optional Face Swap. Its fidelity control lets you balance restoration against identity preservation instead of sending the image through a separate application.
Automatic placementRuns in the post-generation sequence
Adjustable balanceControl restoration versus fidelity
Local workflowReview the finished result in Generation History
This comparison uses genuine output from 65West AI Image Generator (ZT/K2). No fidelity value is claimed because the setting was not recorded with the supplied example.
BEFORE FACE DETAILERAFTER FACE DETAILER
Drag the divider across the face to inspect the genuine result.
Open full size ↗
Inside the applicationActual face-tool controls
Select a source manually or with a prompt face-reference tag, reuse its cached analysis, optionally swap the largest detected face, and tune GFPGAN restoration against identity fidelity—all without leaving the generation workspace.
CACHED IMAGE COMPARISON
See what the LoRA changed.
Keep the test controlled.
Reuse the cached generation state, apply the LoRAs currently selected in the interface, and compare the new result against the baseline. This genuine application example shows the creative influence clearly—without presenting two unrelated generations as a test.
View full size ↗
View full size ↗
864 × 1152 Both original outputsSame source setup Controlled comparisonReal application images No simulated artwork
Open full size ↗
Inside the applicationActual LoRA management controls
Build a local stack, edit strengths directly, use prompt tags, pause selected sources without clearing the setup, and keep comparison results visible in Generation History.
Actual application slideshow viewerGenuine interface capture and locally generated output
A clean black presentation canvas keeps attention on the image while essential slideshow controls remain available along the bottom edge.
THE BORING PARTS MATTER
Built to manage a real local workstation
Generation quality is only part of the experience. 65West also handles the practical details that make a local AI tool comfortable to live with.
Smart model optimization
Unfamiliar configurations are safety-tested, while compatible model switches can reuse the text encoder and unchanged VAE. Sampler and scheduler changes reconfigure the cached pipeline in place.
Memory modes
Auto, an optimized 16 GB path, high-VRAM BF16, CPU offload and sequential CPU offload give you multiple ways to balance speed and memory pressure.
Disk-space control
An optional cache policy—enabled by default—can keep only the active standalone model’s optimized Transformers cache and report reclaimed disk space.
Resource monitor
GPU RAM, system RAM, generation progress and elapsed time stay visible while you work.
Session protection
The app tracks what has and has not been saved and only warns at exit when unsaved Generation History items still exist.
Workspace memory
Live-preview position, layout choices, hidden groups, LoRA states and notes are remembered so the workspace feels like your own.
WHY LOCAL MATTERS
More creating. Less counting.
Cloud image services are convenient, but many tie the workflow to recurring plans, monthly credits or provider limits. A local-first tool shifts the economics: your compatible GPU does the rendering, and the software becomes the workstation around it.
65West local workflowTypical cloud service
Generation cost after activationNo per-image 65West feeOften plan / credit based
Where images are processedYour machineProvider infrastructure
Model / LoRA controlLocal librariesPlatform dependent
Up-front hardwareCompatible GPU neededUsually not needed
Best fitFrequent local creatorsConvenience / occasional use
RECOMMENDED HARDWARE
Give local generation room to work.
For the best Z-Image Turbo and Krea 2 experience, use the recommended configuration below. Lower-spec systems may work with CPU-offload modes, but generation will be slower and available image sizes may be more limited.
Operating systemWindows 11, 64-bit
The application is designed for a current 64-bit Windows 11 desktop with an up-to-date NVIDIA graphics driver.
Graphics cardNVIDIA RTX · 16 GB VRAM
A 16 GB NVIDIA RTX GPU, such as the GeForce RTX 5060 Ti 16 GB or better, is recommended for the optimized 16 GB workflow.
System memory32 GB RAM or more
32 GB is recommended for normal local generation. 64 GB provides additional headroom for CPU offload, larger workflows, and model switching.
Storage100 GB free SSD space
An NVMe SSD is recommended. The supported model families can use roughly 70 GB, with additional space needed for LoRAs, VAEs, caches, upscalers, and generated images.
ProcessorModern 6-core CPU or better
A recent Intel Core or AMD Ryzen processor is recommended. The GPU performs the main generation workload, while the CPU supports loading, offload, and application tasks.
Model compatibilityMatch assets to the family
Use models, VAEs, and LoRAs made for the selected Z-Image Turbo or Krea 2 family. Assets from Stable Diffusion, SDXL, or the other family are not automatically interchangeable.
Important: Protected model access may require accepting the model provider’s terms before setup can download it. Model downloads and generation require compatible third-party licenses, enough disk space, and sufficient local resources.
ONE-TIME PURCHASE
Buy the app. Keep the app.
A straightforward perpetual license fits a local desktop generator better than charging users every month just to keep using their own GPU.
FULL VERSION
65West AI Image Generator (ZT/K2)Personal perpetual license
$39.95one-time purchase
✓ Current Windows application✓ Two active devices✓ No recurring fee to keep the purchased version✓ Full HTML user guide✓ No per-image 65West charge after activation✓ License delivery by email
One-time purchase through Paddle. Applicable taxes are calculated at checkout.
Not ready to buy?Try the complete application with a Demo watermark.
Download DemoNo payment card, generation-count limit, time limit, or intentionally reduced output quality. Demo images contain a translucent tiled 65West watermark. Activating a license removes the watermark from future generations.
SECURE PURCHASE
Purchase through Paddle
Checkout is securely provided by Paddle, the merchant of record. Paddle handles payment processing, applicable taxes, and the purchase receipt.
1
Choose Buy Now
The customer sees the exact product, license type, update period and total price before continuing.
2
Enter delivery email
The email becomes the purchase receipt and license-delivery destination.
3
Secure payment
Paddle processes the payment and applicable tax. 65West does not receive or store your complete card details.
4
Download + activate
Your license key is delivered by email. Enter it in the installed Demo to remove the watermark from future generations—no reinstall is required.
QUESTIONS BEFORE PURCHASE
FAQ
Does 65West AI Image Generator (ZT/K2) charge per image?
No. There is no generation-count limit or per-image 65West charge. The Demo remains fully functional and places a translucent tiled 65West watermark on completed images. A purchased license removes that watermark from future generations.
Which model families does the application support?
The ZT/K2 release supports compatible Z-Image Turbo and Krea 2 models. It applies family-aware defaults and VAE handling when you switch. LoRAs and related assets must be compatible with the active family; Krea 2 does not support negative prompts.
What are the Demo restrictions?
The Demo has no image-count or time limit. Its restriction is a translucent tiled 65West watermark applied to completed, displayed, preserved, and saved images. Activating a license removes the watermark from future generations without requiring reinstallation.
Can I present my generated images as a slideshow?
Yes. The built-in fullscreen slideshow viewer presents session images against a clean black canvas with Pause, optional Randomize, current-image position, filename, and Close controls. A separate actual-size viewer is also available for close inspection.
Does it work like ComfyUI?
It solves a different usability problem. ComfyUI is a highly modular node-based engine; 65West is a focused desktop interface for Z-Image Turbo and Krea 2 workflows, with model-family switching, compatible LoRAs, refinement, face, history, and viewing tools organized into one application.
Can I use Stable Diffusion or SDXL LoRAs?
Not automatically. The user guide specifically warns that LoRAs built for unrelated Stable Diffusion or SDXL UNet architectures are not inherently compatible with Z-Image Turbo.
What does Image Refinement do?
It is a genuine second image-to-image diffusion pass, not a simple final enlargement. It can target detail, lighting, tone and depth through selectable profiles and strengths while reusing the already-loaded compatible model components.
Does the program include face tools?
Yes. Optional Face Swap uses a prompt face-reference tag and can be followed by GFPGAN Face Detailer processing. Users should only use face images they have the right and consent to use.
Does the program upscale the final image?
Yes, optionally. Image Refinement is a second diffusion pass that works on image content, while final upscaling is a separate last-stage process using a compatible local PyTorch super-resolution model. The requested output size can range from 1.5× through 4.0×. Preserve Original-Size Image is disabled by default.
Are models included with the purchase?
Setup can download supported default models after any required provider access has been accepted. The application also supports compatible local model libraries and supported repository identifiers. Third-party models, VAEs, LoRAs, and other assets remain subject to their own licenses and terms.
Can generated images be used commercially?
The software license and the rights to generated output are separate issues. Output rights can depend on the model, LoRA and other assets used, plus applicable law. Customers should review the license terms for every third-party model or asset in their workflow.
65WEST AI IMAGE GENERATOR (ZT/K2)
Turn your Windows PC into a focused multi-model AI image studio.
Choose a model family. Generate. Refine. Compare. Save. Stay in control.