AI Generation

FAQ about AI content generation

Last updated: 2026-11-13

AI Generation FAQ

Common questions and answers about using HoldOrange's content generation features.


Why is my generation taking so long?

Generation time depends on several factors:

  • Model complexity: Higher-quality models (e.g., ultra-HD or cinematic modes) require significantly more processing time than standard models.
  • Resolution: Larger output resolutions (2K, 4K) take longer to render than standard sizes.
  • Server load: During peak hours, the generation queue may be longer. Typical wait times range from 30 seconds to several minutes.
  • Prompt complexity: Highly detailed prompts with many elements, style references, or compositional constraints require more computation.
💡Tip
If your generation is taking longer than 10 minutes, try refreshing the page and checking the Generation History panel. Your task may have already completed. You can also check the platform status page for any ongoing service disruptions.

If generations consistently take too long, consider:

  1. Reducing the output resolution temporarily
  2. Simplifying your prompt
  3. Trying again during off-peak hours (early morning or late evening UTC)

Why did my generation fail?

Generations can fail for several reasons:

  • Content policy violation: Your prompt may contain terms that trigger the platform's safety filters. Review the actual error message and adjust your input accordingly.
  • Server timeout: If the server is under heavy load, long-running tasks may time out. Simply retry your generation.
  • Invalid parameters: Certain combinations of settings (e.g., incompatible aspect ratios with specific models) can cause failures.
  • Insufficient credits: If your credit balance drops to zero mid-generation, the task will be cancelled.
💡Tip
Check the error message in your Generation History — it usually contains a specific error code. You can search this code in our documentation or contact support with it for faster resolution.

Common error codes: | Code | Meaning | Solution | |------|---------|----------| | GEN_TIMEOUT | Server timeout | Retry the generation | | GEN_POLICY | Content policy violation | Revise your prompt | | GEN_PARAMS | Invalid parameters | Check your settings | | GEN_QUOTA | Insufficient credits | Top up your balance |


How do I write better prompts?

Writing effective prompts is key to getting great results. Here are best practices:

  1. Be specific and descriptive: Instead of "a cat", try "a fluffy orange tabby cat sitting on a windowsill, soft afternoon sunlight, photorealistic style".
  2. Specify the style: Mention artistic styles like "watercolor", "cinematic", "photorealistic", "anime", or "oil painting".
  3. Describe composition: Use terms like "close-up", "wide angle", "bird's eye view", or "centered subject".
  4. Include lighting details: "Golden hour lighting", "dramatic shadows", "soft diffused light", or "neon glow".
  5. Use negative prompts: Specify what you don't want to see — e.g., "no text, no watermarks, no blurry elements".
💡Tip
Use the Prompt Assistant feature (available in the generation panel) to get AI-powered suggestions for improving your prompts. It can help expand vague descriptions into detailed, model-friendly prompts.

Example of a weak vs. strong prompt:

  • ❌ Weak: "A landscape"
  • ✅ Strong: "A sweeping mountain landscape at sunset, snow-capped peaks reflecting golden light, a crystal-clear lake in the foreground, pine forests along the shores, dramatic clouds, photorealistic, ultra-detailed, 8K quality"

What image resolutions are supported?

HoldOrange supports a range of output resolutions depending on your subscription tier:

ResolutionAspect RatiosAvailability
512 × 5121:1All plans
768 × 7681:1All plans
1024 × 10241:1All plans
1024 × 57616:9All plans
1152 × 8964:3Pro and above
1536 × 10243:2Pro and above
2048 × 20481:1Enterprise only
💡Tip
For social media content, use preset aspect ratios: select "Instagram Post" (1:1), "Instagram Story" (9:16), "YouTube Thumbnail" (16:9), or "Twitter Header" (3:1) from the quick-select menu to automatically configure dimensions.

Custom resolutions are available on Enterprise plans. Contact your account manager for details.


Can I upscale generated images?

Yes! HoldOrange offers built-in upscaling capabilities:

  1. Open any generated image from your Generation History or Gallery.
  2. Click the Upscale button (magnifying glass icon) in the image toolbar.
  3. Choose your target scale: 2×, 4×, or Custom.
  4. The upscaled image will appear in your history once processing is complete.

Upscaling uses AI-powered super-resolution, so it enhances detail rather than simply stretching pixels. Upscaling costs additional credits based on the scale factor and original image size.

💡Tip
For the best upscaling results, start with the highest quality generation you can. Upscaling a well-generated 1024×1024 image to 4K will yield much better results than upscaling a 512×512 image.

Why does the output look different from what I described?

AI generation is inherently probabilistic — the model interprets your text prompt and produces its best visual interpretation. Differences can occur because:

  • Ambiguous language: Words like "big" or "nice" are subjective. Use precise measurements or comparisons instead.
  • Conflicting descriptions: Asking for "a dark, bright room" confuses the model. Ensure your prompt is internally consistent.
  • Model limitations: Current models may struggle with exact text rendering, precise counting of objects, or very specific spatial arrangements.
  • Seed randomness: Each generation uses a random seed. The same prompt will produce slightly different results each time.

Tips for more consistent results:

  • Use the seed lock feature to reproduce similar outputs across runs.
  • Provide reference images when available to guide the model's interpretation.
  • Break complex scenes into simpler elements and use compositing tools.
💡Tip
If you get a result that's close but not perfect, use the Variation feature to generate subtle modifications of that specific output rather than starting from scratch. This preserves the overall composition while refining details.

How many generations can I run simultaneously?

Concurrent generation limits depend on your subscription tier:

PlanConcurrent GenerationsQueue Limit
Free13
Basic210
Pro525
Enterprise10+Unlimited

When you exceed your concurrent limit, additional generations are placed in a queue and processed in order. You can view your queue status in the Generation History panel.

💡Tip
Batch generation mode (available on Pro plans and above) lets you submit multiple prompt variations at once and have them processed sequentially, which is more efficient than manually queuing individual generations.

What's the difference between generation modes?

HoldOrange offers several generation modes tailored to different use cases:

  • Standard Mode: Best for general-purpose image creation. Balanced speed and quality. Ideal for quick iterations and concept exploration.
  • Quality Mode: Produces higher-fidelity outputs with more detail and coherence. Takes longer and costs more credits. Best for final deliverables.
  • Creative Mode: Applies more artistic interpretation to your prompts. Results may be more surprising or abstract. Great for brainstorming and inspiration.
  • Precise Mode: Follows your prompt more literally with less creative liberty. Best when you need exact adherence to your description.
  • Video Mode: Generates short video clips from text or image inputs. Supports 3–10 second clips with configurable frame rates.
💡Tip
Not sure which mode to use? Start with Standard Mode for quick previews, then switch to Quality Mode once you're happy with the composition and prompt. This canvas saves credits during the iteration phase.

Can I use my own reference images?

Yes, HoldOrange supports several reference image canvases:

  1. Image-to-Image (img2img): Upload a reference image and combine it with a text prompt. The model uses your image as a compositional guide. Adjust the influence strength slider to control how closely the output follows your reference (0% = ignore reference, 100% = closely match).
  1. Style Reference: Upload an image purely for its artistic style. The model will apply the color palette, texture, and aesthetic of your reference to the generated content while following your text prompt for subject matter.
  1. ControlNet / Structural Reference: Upload sketches, depth maps, or edge-detected images to precisely control the spatial layout of your generation.
  1. Face / Character Reference: Upload a portrait to maintain consistent character appearance across multiple generations.
💡Tip
Supported upload formats include JPG, PNG, and WebP. For best results, use reference images with a resolution of at least 512×512 and avoid heavily compressed or watermarked sources. The maximum upload size is 10 MB per image.