Common Mistakes to Avoid

Learn from the most common mistakes Armox users make. Avoiding these pitfalls will save you time, credits, and frustration.

Prompt Mistakes

Mistake 1: Too Vague

❌ Bad:

A nice picture of a person

✅ Good:

Professional headshot of a confident businesswoman, 
modern office background, soft natural lighting, 
warm and approachable expression, 
shot on 85mm lens, shallow depth of field

Why it matters: Vague prompts give unpredictable results. The AI fills in details randomly.

Mistake 2: Conflicting Instructions

❌ Bad:

Dark moody lighting, bright and cheerful atmosphere

✅ Good:

Moody lighting with warm accent highlights, 
sophisticated atmosphere

Why it matters: Conflicting instructions confuse the AI and produce inconsistent results.

Mistake 3: Too Many Details

❌ Bad:

A woman with long blonde hair and blue eyes wearing a red 
dress with yellow flowers and green shoes standing in a 
purple room with orange furniture next to a pink dog...

✅ Good:

Elegant woman in floral dress, 
modern minimalist interior, 
soft natural lighting, 
fashion editorial style

Why it matters: Overloaded prompts often result in chaotic images where the AI can't prioritize.

Mistake 4: Not Specifying Style

❌ Bad:

A coffee cup on a table

✅ Good:

Artisan coffee cup on wooden table, 
lifestyle photography, warm morning light, 
Instagram aesthetic, shallow depth of field

Why it matters: Without style guidance, you get generic, stock-photo-like results.

Model Selection Mistakes

Mistake 5: Using Premium for Testing

❌ Bad: Testing prompt iterations with Flux 2 Pro (120 credits each)

✅ Good: Testing with Qwen Image (20 credits), then using Flux 2 Pro for final

Why it matters: You'll burn through credits before finding the right prompt.

Mistake 6: Wrong Model for the Task

❌ Bad: Using SDXL for photorealistic product photography

✅ Good: Using Flux for photorealistic, SDXL for artistic styles

Why it matters: Each model has strengths. Matching model to task gives better results.

Mistake 7: Ignoring Video Model Costs

❌ Bad: Iterating with Veo 3.1 (4,000 credits each)

✅ Good: Testing with Wan 2.5 Fast (60 credits), final with Veo

Why it matters: Video is expensive. One Veo generation = 66 Wan Fast tests.

Workflow Mistakes

Mistake 8: Not Testing Incrementally

❌ Bad: Building a 10-node workflow and running it all at once

✅ Good: Testing each node individually, then running the full workflow

Why it matters: If something fails, you've wasted credits on all subsequent nodes.

Mistake 9: Upscaling Too Early

❌ Bad:

Generate → Upscale → Edit → Upscale again

✅ Good:

Generate → Edit → Final check → Upscale once

Why it matters: Upscaling costs 1,000 credits. Only do it on final images.

Mistake 10: Not Using Reference Images

❌ Bad: Describing everything in text for consistency

✅ Good: Using reference images with Flux Kontext for consistent results

Why it matters: Reference images provide much more consistent results than text alone.

Aspect Ratio Mistakes

Mistake 11: Wrong Ratio for Platform

❌ Bad: Generating 16:9 for Instagram Stories

✅ Good: Generating 9:16 for Stories, 1:1 or 4:5 for feed

Why it matters: Wrong ratios mean awkward cropping or wasted generation.

Mistake 12: Not Considering Composition

❌ Bad: Generating square image with subject at edge

✅ Good: Composing with subject centered or following rule of thirds

Why it matters: Poor composition leads to awkward cropping or unusable images.

Video Mistakes

Mistake 13: Too Complex Motion

❌ Bad:

Person runs, jumps, spins, catches a ball, 
throws it, and does a backflip

✅ Good:

Person slowly turns head toward camera, 
gentle smile forming, 
subtle movement

Why it matters: AI video works best with simple, clear motions.

Mistake 14: Ignoring Duration Limits

❌ Bad: Expecting 60-second video from a 5-second model

✅ Good: Generating multiple 5-second clips and combining

Why it matters: Each model has duration limits. Plan accordingly.

Audio Mistakes

Mistake 15: Poor Voice Clone Samples

❌ Bad: Using noisy, short, or multi-speaker audio for cloning

✅ Good: Clean, 20-30 second, single-speaker sample

Why it matters: Bad samples = bad clones. Quality in = quality out.

Mistake 16: Mismatched Audio Duration

❌ Bad: Generating 30-second music for 10-second video

✅ Good: Matching audio duration to video length

Why it matters: Wastes credits and requires editing anyway.

Organization Mistakes

Mistake 17: Not Saving Successful Prompts

❌ Bad: Recreating prompts from memory each time

✅ Good: Saving successful prompts in a document or App

Why it matters: You'll waste time and credits rediscovering what works.

Mistake 18: Messy Canvas Workflows

❌ Bad: Nodes scattered randomly, unclear connections

✅ Good: Left-to-right flow, organized layout, clear naming

Why it matters: Hard to debug, easy to make connection errors.

Mistake 19: Not Using Brains Properly

❌ Bad: All projects in one Brain

✅ Good: Separate Brains for different clients/projects

Why it matters: Keeps content organized, easier to find assets.

Quick Checklist

Before generating, ask:

• Is my prompt specific enough?
• Am I using the right model for this task?
• Is this the right aspect ratio?
• Am I using a budget model for testing?
• Have I tested this workflow step by step?
• Do I have a reference image if needed?
• Is the motion/audio duration appropriate?
• Should I upscale this, or is it just for web?

Next Steps

• Credit Optimization — Save credits
• Keyboard Shortcuts — Work faster
• Prompt Engineering — Write better prompts